Upload 108 files

README.md

@@ -1,13 +1,73 @@
 ---
-title: ReCEP
-emoji: 🏃
-colorFrom: pink
-colorTo: red
+title: B-cell Epitope Prediction Server
+emoji: 🧬
+colorFrom: blue
+colorTo: purple
 sdk: gradio
-sdk_version: 5.36.2
+sdk_version: 4.44.1
 app_file: app.py
 pinned: false
 license: mit
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🧬 B-cell Epitope Prediction Server
+
+A web-based interface for B-cell epitope prediction using the ReCEP model.
+
+## 📋 How to Use
+
+### 1. Input Protein Structure
+
+Choose one of two input methods:
+
+#### Option A: PDB ID
+- Enter a 4-character PDB ID (e.g., "5I9Q")
+- Specify the chain ID (e.g., "A")
+
+#### Option B: Upload PDB File
+- Upload a PDB structure file (.pdb or .ent format)
+- Optionally specify a custom PDB ID
+- Specify the chain ID
+
+### 2. Configure Prediction Parameters
+
+#### Basic Parameters:
+- **Chain ID**: Target protein chain (default: A)
+
+#### Advanced Parameters (Optional):
+- **Radius**: Spherical region radius in Ångstroms (default: 19.0)
+- **Top-k Regions**: Number of top regions to analyze (default: 7)
+- **Encoder**: Protein encoder type (ESM-C only now)
+- **Device Configuration**: CPU or GPU processing (CPU Only now)
+- **Threshold**: Custom prediction threshold (leave empty for auto, required)
+
+### 3. View Results
+
+The application provides:
+
+#### Prediction Summary
+- Protein information (PDB ID, chain, length, sequence)
+- Prediction statistics (epitope count, coverage rate, etc.)
+- Top-k region centers
+- Predicted epitope residues
+- Binding region residues
+
+#### Download Options
+- **JSON Results**: Complete prediction data with metadata
+- **CSV Results**: Residue-level predictions for analysis
+- **3D Visualization**: Interactive HTML file with 3Dmol.js viewer
+
+### 4. 3D Visualization
+
+The downloadable HTML file includes:
+- **Display Modes**: 
+  - Predicted Epitopes: Highlight predicted epitope residues
+  - Probability Gradient: Color residues by prediction confidence
+- **Representation Styles**: Cartoon, Surface, Stick, Sphere
+- **Interactive Controls**: Rotate, zoom, pan, reset view, save image
+
+## 📜 License
+
+This project is licensed under the MIT License - see the LICENSE file for details.
+
+**Note**: This is a research tool for B-cell epitope prediction. Results should be validated through experimental methods for clinical or commercial applications.

app.py

@@ -0,0 +1,1188 @@
+import gradio as gr
+import os
+import json
+import tempfile
+import traceback
+import numpy as np
+import pandas as pd
+from pathlib import Path
+from typing import Optional, Tuple, Dict, Any
+import torch
+import time
+import io
+import base64
+import zipfile
+from datetime import datetime
+
+# Dynamic installation of PyTorch Geometric dependencies
+def install_torch_geometric_deps():
+    """Install PyTorch Geometric dependencies at runtime to avoid compilation issues during Hugging Face Spaces build"""
+    import subprocess
+    import sys
+    
+    # Check if torch-scatter is already installed
+    try:
+        import torch_scatter
+        print("✅ torch-scatter already installed")
+        return True
+    except ImportError:
+        print("🔄 Installing torch-scatter and related packages...")
+        
+        # Get PyTorch version and CUDA info
+        torch_version = torch.__version__
+        torch_version_str = '+'.join(torch_version.split('+')[:1])  # Remove CUDA info
+        
+        # Use PyTorch Geometric official recommended installation method
+        try:
+            # For CPU version, use official CPU wheel
+            pip_cmd = [
+                sys.executable, "-m", "pip", "install", 
+                "torch-scatter", "torch-sparse", "torch-cluster", "torch-spline-conv",
+                "-f", f"https://data.pyg.org/whl/torch-{torch_version_str}+cpu.html",
+                "--no-cache-dir"
+            ]
+            
+            print(f"Running: {' '.join(pip_cmd)}")
+            result = subprocess.run(pip_cmd, capture_output=True, text=True, timeout=300)
+            
+            if result.returncode == 0:
+                print("✅ Successfully installed torch-scatter and related packages")
+                return True
+            else:
+                print(f"❌ Failed to install packages: {result.stderr}")
+                # Try simplified installation method
+                try:
+                    simple_cmd = [sys.executable, "-m", "pip", "install", "torch-scatter", "--no-cache-dir"]
+                    result = subprocess.run(simple_cmd, capture_output=True, text=True, timeout=300)
+                    if result.returncode == 0:
+                        print("✅ Successfully installed torch-scatter with simple method")
+                        return True
+                    else:
+                        print(f"❌ Simple install also failed: {result.stderr}")
+                        return False
+                except Exception as e:
+                    print(f"❌ Exception during simple install: {e}")
+                    return False
+                    
+        except subprocess.TimeoutExpired:
+            print("❌ Installation timeout - packages may not be available")
+            return False
+        except Exception as e:
+            print(f"❌ Exception during installation: {e}")
+            return False
+
+# Try to install PyTorch Geometric dependencies
+deps_installed = install_torch_geometric_deps()
+
+if not deps_installed:
+    print("⚠️ Warning: PyTorch Geometric dependencies not installed. Some features may not work.")
+    print("The application will try to continue with limited functionality.")
+
+# Set up paths and imports for different deployment environments
+import sys
+BASE_DIR = Path(__file__).parent
+
+# Smart import handling for different environments
+def setup_imports():
+    """Smart import setup for different deployment environments"""
+    global AntigenChain, PROJECT_BASE_DIR
+    
+    # Method 1: Try importing from src directory (local development)
+    if (BASE_DIR / "src").exists():
+        sys.path.insert(0, str(BASE_DIR))
+        try:
+            from src.bce.antigen.antigen import AntigenChain
+            from src.bce.utils.constants import BASE_DIR as PROJECT_BASE_DIR
+            print("✅ Successfully imported from src/ directory")
+            return True
+        except ImportError as e:
+            print(f"❌ Failed to import from src/: {e}")
+    
+    # Method 2: Try adding src to path and direct import (Hugging Face Spaces)
+    src_path = BASE_DIR / "src"
+    if src_path.exists():
+        sys.path.insert(0, str(src_path))
+        try:
+            from bce.antigen.antigen import AntigenChain
+            from bce.utils.constants import BASE_DIR as PROJECT_BASE_DIR
+            print("✅ Successfully imported from src/ added to path")
+            return True
+        except ImportError as e:
+            print(f"❌ Failed to import with src/ in path: {e}")
+    
+    # Method 3: Try direct import (if package is installed)
+    try:
+        from bce.antigen.antigen import AntigenChain
+        from bce.utils.constants import BASE_DIR as PROJECT_BASE_DIR
+        print("✅ Successfully imported from installed package")
+        return True
+    except ImportError as e:
+        print(f"❌ Failed to import from installed package: {e}")
+    
+    # If all methods fail, use default settings
+    print("⚠️ All import methods failed, using fallback settings")
+    PROJECT_BASE_DIR = BASE_DIR
+    return False
+
+# Execute import setup
+import_success = setup_imports()
+
+if not import_success:
+    print("❌ Critical: Could not import BCE modules. Please check the file structure.")
+    print("Expected structure:")
+    print("- src/bce/antigen/antigen.py")
+    print("- src/bce/utils/constants.py")
+    print("- src/bce/model/ReCEP.py")
+    print("- src/bce/data/utils.py")
+    sys.exit(1)
+
+# Configuration
+DEFAULT_MODEL_PATH = os.getenv("BCE_MODEL_PATH", str(PROJECT_BASE_DIR / "models" / "ReCEP" / "20250626_110438" / "best_mcc_model.bin"))
+ESM_TOKEN = os.getenv("ESM_TOKEN", "1mzAo8l1uxaU8UfVcGgV7B")
+
+# PDB data directory
+PDB_DATA_DIR = PROJECT_BASE_DIR / "data" / "pdb"
+PDB_DATA_DIR.mkdir(parents=True, exist_ok=True)
+
+def validate_pdb_id(pdb_id: str) -> bool:
+    """Validate PDB ID format"""
+    if not pdb_id or len(pdb_id) != 4:
+        return False
+    return pdb_id.isalnum()
+
+def validate_chain_id(chain_id: str) -> bool:
+    """Validate chain ID format"""
+    if not chain_id or len(chain_id) != 1:
+        return False
+    return chain_id.isalnum()
+
+def create_pdb_visualization_html(pdb_data: str, predicted_epitopes: list, 
+                                 predictions: dict, protein_id: str, top_k_regions: list = None) -> str:
+    """Create HTML with 3Dmol.js visualization compatible with Gradio - enhanced version with more features"""
+    
+    # Prepare data for JavaScript
+    epitope_residues = predicted_epitopes
+    
+    # Process top_k_regions for visualization
+    processed_regions = []
+    if top_k_regions:
+        for i, region in enumerate(top_k_regions):
+            if isinstance(region, dict):
+                processed_regions.append({
+                    'center_idx': region.get('center_idx', 0),
+                    'center_residue': region.get('center_residue', region.get('center_idx', 0)),
+                    'covered_residues': region.get('covered_residues', region.get('covered_indices', [])),
+                    'radius': 19.0,  # Default radius
+                    'predicted_value': region.get('graph_pred', 0.0)
+                })
+    
+    # Create a unique ID for this visualization to avoid conflicts
+    import uuid
+    viewer_id = f"viewer_{uuid.uuid4().hex[:8]}"
+    
+    html_content = f"""
+    <div style="width: 100%; height: 600px; border: 1px solid #ddd; border-radius: 8px; overflow: hidden;">
+        <div style="padding: 10px; background: #f8f9fa; border-bottom: 1px solid #ddd;">
+            <h3 style="margin: 0 0 10px 0; color: #333;">3D Structure Visualization - {protein_id}</h3>
+            <div style="display: flex; gap: 15px; align-items: center; flex-wrap: wrap;">
+                <div>
+                    <label style="font-weight: bold; margin-right: 5px;">Display Mode:</label>
+                    <select id="vizMode_{viewer_id}" onchange="updateVisualization_{viewer_id}()" style="padding: 4px;">
+                        <option value="prediction">Predicted Epitopes</option>
+                        <option value="probability">Probability Gradient</option>
+                        <option value="regions">Top-k Regions</option>
+                    </select>
+                </div>
+                <div>
+                    <label style="font-weight: bold; margin-right: 5px;">Style:</label>
+                    <select id="vizStyle_{viewer_id}" onchange="updateVisualization_{viewer_id}()" style="padding: 4px;">
+                        <option value="cartoon">Cartoon</option>
+                        <option value="surface">Surface</option>
+                        <option value="stick">Stick</option>
+                        <option value="sphere">Sphere</option>
+                    </select>
+                </div>
+                <div>
+                    <label style="font-weight: bold; margin-right: 5px;">
+                        <input type="checkbox" id="showSpheres_{viewer_id}" onchange="updateVisualization_{viewer_id}()" style="margin-right: 3px;"> Show Spheres
+                    </label>
+                </div>
+                <div>
+                    <label style="font-weight: bold; margin-right: 5px;">Sphere Display:</label>
+                    <select id="sphereCount_{viewer_id}" onchange="handleSphereCountChange_{viewer_id}()" style="padding: 4px;">
+                        <option value="1">Top 1</option>
+                        <option value="2">Top 2</option>
+                        <option value="3">Top 3</option>
+                        <option value="4">Top 4</option>
+                        <option value="5" selected>Top 5</option>
+                        <option value="6">Top 6</option>
+                        <option value="7">Top 7</option>
+                        <option value="all">All Spheres</option>
+                        <option value="custom">Custom Selection</option>
+                    </select>
+                </div>
+                <div id="customSphereSelection_{viewer_id}" style="display: none; margin-top: 10px; padding: 10px; background: #f9f9f9; border-radius: 5px; max-height: 120px; overflow-y: auto;">
+                    <label style="font-weight: bold; margin-bottom: 5px; display: block;">Select Spheres to Display:</label>
+                    <div id="sphereCheckboxes_{viewer_id}" style="display: flex; flex-wrap: wrap; gap: 8px; max-height: 80px; overflow-y: auto;">
+                        <!-- Checkboxes will be dynamically generated -->
+                    </div>
+                </div>
+                <div>
+                    <button onclick="resetView_{viewer_id}()" style="padding: 4px 8px; margin-right: 5px;">Reset View</button>
+                    <button onclick="saveImage_{viewer_id}()" style="padding: 4px 8px;">Save Image</button>
+                </div>
+            </div>
+        </div>
+        <div id="{viewer_id}" style="width: 100%; height: 520px; min-height: 400px; position: relative; background: #f0f0f0;">
+            <div style="position: absolute; top: 50%; left: 50%; transform: translate(-50%, -50%); text-align: center;">
+                <p id="status_{viewer_id}" style="color: #666;">Loading 3Dmol.js...</p>
+            </div>
+        </div>
+    </div>
+    
+    <script src="https://unpkg.com/[email protected]/build/3Dmol-min.js"></script>
+    <script>
+        // Global variables for this viewer instance
+        window.viewer_{viewer_id} = null;
+        window.pdbData_{viewer_id} = `{pdb_data}`;
+        window.predictedEpitopes_{viewer_id} = {json.dumps(epitope_residues)};
+        window.predictions_{viewer_id} = {json.dumps(predictions)};
+        window.topKRegions_{viewer_id} = {json.dumps(processed_regions)};
+        
+        // Wait for 3Dmol to be available with timeout
+        function wait3Dmol_{viewer_id}(attempts = 0) {{
+            if (typeof $3Dmol !== 'undefined') {{
+                console.log('3Dmol.js loaded successfully for {viewer_id}');
+                document.getElementById('status_{viewer_id}').textContent = 'Initializing 3D viewer...';
+                setTimeout(() => initializeViewer_{viewer_id}(), 100);
+            }} else if (attempts < 50) {{ // 5 second timeout
+                console.log(`Waiting for 3Dmol.js... attempt ${{attempts + 1}}`);
+                setTimeout(() => wait3Dmol_{viewer_id}(attempts + 1), 100);
+            }} else {{
+                console.error('Failed to load 3Dmol.js after 5 seconds');
+                document.getElementById('status_{viewer_id}').textContent = 'Failed to load 3Dmol.js. Please refresh the page.';
+                document.getElementById('status_{viewer_id}').style.color = 'red';
+            }}
+        }}
+        
+        function initializeViewer_{viewer_id}() {{
+            try {{
+                const element = document.getElementById('{viewer_id}');
+                if (!element) {{
+                    console.error('Viewer element not found: {viewer_id}');
+                    return;
+                }}
+                
+                document.getElementById('status_{viewer_id}').textContent = 'Creating viewer...';
+                
+                window.viewer_{viewer_id} = $3Dmol.createViewer(element, {{
+                    defaultcolors: $3Dmol.rasmolElementColors
+                }});
+                
+                document.getElementById('status_{viewer_id}').textContent = 'Loading structure...';
+                
+                window.viewer_{viewer_id}.addModel(window.pdbData_{viewer_id}, 'pdb');
+                
+                // Hide status message
+                const statusEl = document.getElementById('status_{viewer_id}');
+                if (statusEl) statusEl.style.display = 'none';
+                
+                updateVisualization_{viewer_id}();
+                
+                // Initialize sphere checkboxes if data is available
+                if (window.topKRegions_{viewer_id} && window.topKRegions_{viewer_id}.length > 0) {{
+                    generateSphereCheckboxes_{viewer_id}();
+                }}
+                
+                console.log('3D viewer initialized successfully for {viewer_id}');
+            }} catch (error) {{
+                console.error('Error initializing 3D viewer:', error);
+                const statusEl = document.getElementById('status_{viewer_id}');
+                if (statusEl) {{
+                    statusEl.textContent = 'Error loading 3D viewer: ' + error.message;
+                    statusEl.style.color = 'red';
+                }}
+            }}
+        }}
+        
+        function updateVisualization_{viewer_id}() {{
+            if (!window.viewer_{viewer_id}) return;
+            
+            try {{
+                const mode = document.getElementById('vizMode_{viewer_id}').value;
+                const style = document.getElementById('vizStyle_{viewer_id}').value;
+                const showSpheres = document.getElementById('showSpheres_{viewer_id}').checked;
+                
+                // Clear everything
+                window.viewer_{viewer_id}.removeAllShapes();
+                window.viewer_{viewer_id}.removeAllSurfaces();
+                window.viewer_{viewer_id}.setStyle({{}}, {{}});
+                
+                // Base style
+                const baseStyle = {{}};
+                if (style === 'surface') {{
+                    baseStyle['cartoon'] = {{ hidden: true }};
+                }} else {{
+                    baseStyle[style] = {{ color: '#e6e6f7' }};
+                }}
+                window.viewer_{viewer_id}.setStyle({{}}, baseStyle);
+                
+                if (mode === 'prediction') {{
+                    // Highlight predicted epitopes
+                    if (window.predictedEpitopes_{viewer_id}.length > 0 && style !== 'surface') {{
+                        const epitopeStyle = {{}};
+                        epitopeStyle[style] = {{ color: '#9C6ADE' }};
+                        window.viewer_{viewer_id}.setStyle({{ resi: window.predictedEpitopes_{viewer_id} }}, epitopeStyle);
+                    }}
+                    
+                    // Add surface for epitopes if surface mode
+                    if (style === 'surface') {{
+                        window.viewer_{viewer_id}.addSurface($3Dmol.SurfaceType.VDW, {{
+                            opacity: 0.8,
+                            color: '#e6e6f7'
+                        }});
+                        
+                        if (window.predictedEpitopes_{viewer_id}.length > 0) {{
+                            window.viewer_{viewer_id}.addSurface($3Dmol.SurfaceType.VDW, {{
+                                opacity: 1.0,
+                                color: '#9C6ADE'
+                            }}, {{ resi: window.predictedEpitopes_{viewer_id} }});
+                        }}
+                    }}
+                }} else if (mode === 'probability') {{
+                    // Color by probability scores
+                    if (window.predictions_{viewer_id} && Object.keys(window.predictions_{viewer_id}).length > 0) {{
+                        const allProbs = Object.values(window.predictions_{viewer_id}).filter(p => p !== undefined);
+                        const minProb = Math.min(...allProbs);
+                        const maxProb = Math.max(...allProbs);
+                        
+                        Object.entries(window.predictions_{viewer_id}).forEach(([resnum, score]) => {{
+                            const normalizedProb = maxProb > minProb ? (score - minProb) / (maxProb - minProb) : 0.5;
+                            const color = interpolateColor('#E6F3FF', '#DC143C', normalizedProb);
+                            const probStyle = {{}};
+                            if (style !== 'surface') {{
+                                probStyle[style] = {{ color: color }};
+                                window.viewer_{viewer_id}.setStyle({{ resi: parseInt(resnum) }}, probStyle);
+                            }}
+                        }});
+                        
+                        if (style === 'surface') {{
+                            window.viewer_{viewer_id}.addSurface($3Dmol.SurfaceType.VDW, {{
+                                opacity: 0.8,
+                                color: '#e6e6f7'
+                            }});
+                            
+                            Object.entries(window.predictions_{viewer_id}).forEach(([resnum, score]) => {{
+                                const normalizedProb = maxProb > minProb ? (score - minProb) / (maxProb - minProb) : 0.5;
+                                const color = interpolateColor('#E6F3FF', '#DC143C', normalizedProb);
+                                window.viewer_{viewer_id}.addSurface($3Dmol.SurfaceType.VDW, {{
+                                    opacity: 1.0,
+                                    color: color
+                                }}, {{ resi: parseInt(resnum) }});
+                            }});
+                        }}
+                    }}
+                }} else if (mode === 'regions') {{
+                    // Color top-k regions
+                    const colors = ['#FF6B6B', '#96CEB4', '#4ECDC4', '#45B7D1', '#FFEAA7', '#DDA0DD', '#87CEEB'];
+                    
+                    if (window.topKRegions_{viewer_id} && window.topKRegions_{viewer_id}.length > 0) {{
+                        window.topKRegions_{viewer_id}.forEach((region, index) => {{
+                            const color = colors[index % colors.length];
+                            const regionStyle = {{}};
+                            if (style !== 'surface') {{
+                                regionStyle[style] = {{ color: color }};
+                                window.viewer_{viewer_id}.setStyle({{ resi: region.covered_residues }}, regionStyle);
+                            }}
+                        }});
+                        
+                        if (style === 'surface') {{
+                            window.viewer_{viewer_id}.addSurface($3Dmol.SurfaceType.VDW, {{
+                                opacity: 0.8,
+                                color: '#e6e6f7'
+                            }});
+                            
+                            window.topKRegions_{viewer_id}.forEach((region, index) => {{
+                                const color = colors[index % colors.length];
+                                window.viewer_{viewer_id}.addSurface($3Dmol.SurfaceType.VDW, {{
+                                    opacity: 1.0,
+                                    color: color
+                                }}, {{ resi: region.covered_residues }});
+                            }});
+                        }}
+                    }}
+                }}
+                
+                // Add spheres if requested
+                if (showSpheres && window.topKRegions_{viewer_id} && window.topKRegions_{viewer_id}.length > 0) {{
+                    const colors = ['#FF6B6B', '#96CEB4', '#4ECDC4', '#45B7D1', '#FFEAA7', '#DDA0DD', '#87CEEB'];
+                    const sphereCount = document.getElementById('sphereCount_{viewer_id}').value;
+                    
+                    // Determine which spheres to show
+                    let spheresToShow = [];
+                    if (sphereCount === 'custom') {{
+                        const selectedIndices = getSelectedSphereIndices_{viewer_id}();
+                        spheresToShow = selectedIndices.map(idx => ({{ region: window.topKRegions_{viewer_id}[idx], index: idx }}));
+                    }} else {{
+                        let numSpheres = sphereCount === 'all' ? window.topKRegions_{viewer_id}.length : parseInt(sphereCount);
+                        numSpheres = Math.min(numSpheres, window.topKRegions_{viewer_id}.length);
+                        spheresToShow = window.topKRegions_{viewer_id}.slice(0, numSpheres).map((region, index) => ({{ region, index }}));
+                    }}
+                    
+                    spheresToShow.forEach(({{ region, index }}) => {{
+                        const color = colors[index % colors.length];
+                        const centerResidues = window.viewer_{viewer_id}.getModel(0).selectedAtoms({{
+                            resi: region.center_residue,
+                            atom: 'CA'
+                        }});
+                        
+                        if (centerResidues.length > 0) {{
+                            const centerAtom = centerResidues[0];
+                            const centerCoords = {{ x: centerAtom.x, y: centerAtom.y, z: centerAtom.z }};
+                            
+                            // Add wireframe sphere
+                            window.viewer_{viewer_id}.addSphere({{
+                                center: centerCoords,
+                                radius: region.radius,
+                                color: color,
+                                wireframe: true,
+                                linewidth: 2.0
+                            }});
+                            
+                            // Add center point
+                            window.viewer_{viewer_id}.addSphere({{
+                                center: centerCoords,
+                                radius: 0.7,
+                                color: '#FFD700',
+                                wireframe: false
+                            }});
+                        }}
+                    }});
+                }}
+                
+                window.viewer_{viewer_id}.zoomTo();
+                window.viewer_{viewer_id}.render();
+            }} catch (error) {{
+                console.error('Error updating visualization:', error);
+            }}
+        }}
+        
+        // Color interpolation helper functions
+        function interpolateColor(color1, color2, factor) {{
+            const c1 = hexToRgb(color1);
+            const c2 = hexToRgb(color2);
+            
+            const r = Math.round(c1.r + factor * (c2.r - c1.r));
+            const g = Math.round(c1.g + factor * (c2.g - c1.g));
+            const b = Math.round(c1.b + factor * (c2.b - c1.b));
+            
+            return rgbToHex(r, g, b);
+        }}
+        
+        function hexToRgb(hex) {{
+            const result = /^#?([a-f\d]{{2}})([a-f\d]{{2}})([a-f\d]{{2}})$/i.exec(hex);
+            return result ? {{
+                r: parseInt(result[1], 16),
+                g: parseInt(result[2], 16),
+                b: parseInt(result[3], 16)
+            }} : null;
+        }}
+        
+        function rgbToHex(r, g, b) {{
+            return "#" + ((1 << 24) + (r << 16) + (g << 8) + b).toString(16).slice(1);
+        }}
+        
+        function resetView_{viewer_id}() {{
+            if (window.viewer_{viewer_id}) {{
+                window.viewer_{viewer_id}.zoomTo();
+                window.viewer_{viewer_id}.render();
+            }}
+        }}
+        
+        function saveImage_{viewer_id}() {{
+            if (window.viewer_{viewer_id}) {{
+                window.viewer_{viewer_id}.pngURI(function(uri) {{
+                    const link = document.createElement('a');
+                    link.href = uri;
+                    link.download = '{protein_id}_structure.png';
+                    link.click();
+                }});
+            }}
+        }}
+        
+        // Handle sphere count selection change
+        function handleSphereCountChange_{viewer_id}() {{
+            const sphereCount = document.getElementById('sphereCount_{viewer_id}').value;
+            const customSelectionDiv = document.getElementById('customSphereSelection_{viewer_id}');
+            
+            if (sphereCount === 'custom') {{
+                customSelectionDiv.style.display = 'block';
+                generateSphereCheckboxes_{viewer_id}();
+            }} else {{
+                customSelectionDiv.style.display = 'none';
+            }}
+            
+            updateVisualization_{viewer_id}();
+        }}
+        
+        // Generate sphere checkboxes for custom selection
+        function generateSphereCheckboxes_{viewer_id}() {{
+            if (!window.topKRegions_{viewer_id} || window.topKRegions_{viewer_id}.length === 0) {{
+                return;
+            }}
+            
+            const regions = window.topKRegions_{viewer_id};
+            const container = document.getElementById('sphereCheckboxes_{viewer_id}');
+            container.innerHTML = '';
+            
+            regions.forEach((region, index) => {{
+                const sphereNum = index + 1;
+                const checkboxId = `sphere_{{sphereNum}}_{viewer_id}`;
+                const colors = ['#FF6B6B', '#96CEB4', '#4ECDC4', '#45B7D1', '#FFEAA7', '#DDA0DD', '#87CEEB'];
+                const sphereColor = colors[index % colors.length];
+                
+                const checkboxContainer = document.createElement('div');
+                checkboxContainer.style.cssText = `
+                    display: flex;
+                    align-items: center;
+                    padding: 5px 10px;
+                    border: 1px solid #ddd;
+                    border-radius: 4px;
+                    background: white;
+                    cursor: pointer;
+                    user-select: none;
+                `;
+                checkboxContainer.setAttribute('data-sphere', sphereNum);
+                
+                const checkbox = document.createElement('input');
+                checkbox.type = 'checkbox';
+                checkbox.id = checkboxId;
+                checkbox.checked = sphereNum <= 5; // Default: show first 5
+                checkbox.style.marginRight = '5px';
+                
+                const colorBox = document.createElement('div');
+                colorBox.style.cssText = `
+                    width: 16px;
+                    height: 16px;
+                    background-color: ${{sphereColor}};
+                    border: 1px solid #333;
+                    border-radius: 2px;
+                    margin-right: 5px;
+                `;
+                
+                const label = document.createElement('label');
+                label.setAttribute('for', checkboxId);
+                label.textContent = `Sphere ${{sphereNum}} (R${{region.center_residue}})`;
+                label.style.cursor = 'pointer';
+                label.style.fontSize = '14px';
+                
+                checkboxContainer.appendChild(checkbox);
+                checkboxContainer.appendChild(colorBox);
+                checkboxContainer.appendChild(label);
+                container.appendChild(checkboxContainer);
+                
+                // Add click handler
+                checkboxContainer.addEventListener('click', function(e) {{
+                    if (e.target.type !== 'checkbox') {{
+                        checkbox.checked = !checkbox.checked;
+                    }}
+                    
+                    if (checkbox.checked) {{
+                        checkboxContainer.style.backgroundColor = '#f0f8ff';
+                        checkboxContainer.style.borderColor = '#4a90e2';
+                    }} else {{
+                        checkboxContainer.style.backgroundColor = 'white';
+                        checkboxContainer.style.borderColor = '#ddd';
+                    }}
+                    
+                    updateVisualization_{viewer_id}();
+                }});
+                
+                // Initialize visual state
+                if (checkbox.checked) {{
+                    checkboxContainer.style.backgroundColor = '#f0f8ff';
+                    checkboxContainer.style.borderColor = '#4a90e2';
+                }}
+            }});
+        }}
+        
+        // Get selected sphere indices for custom mode
+        function getSelectedSphereIndices_{viewer_id}() {{
+            const selected = [];
+            const checkboxes = document.querySelectorAll('#sphereCheckboxes_{viewer_id} input[type="checkbox"]:checked');
+            checkboxes.forEach(function(checkbox) {{
+                // Get sphere number from the data-sphere attribute of the container
+                const container = checkbox.closest('[data-sphere]');
+                if (container) {{
+                    const sphereNum = parseInt(container.getAttribute('data-sphere'));
+                    selected.push(sphereNum - 1); // Convert to 0-based index
+                }}
+            }});
+            return selected;
+        }}
+        
+        // Start initialization
+        wait3Dmol_{viewer_id}();
+    </script>
+    """
+    
+    return html_content
+
+def predict_epitopes(pdb_id: str, pdb_file, chain_id: str, radius: float, k: int, 
+                    encoder: str, device_config: str, use_threshold: bool, threshold: float,
+                    auto_cleanup: bool, progress: gr.Progress = None) -> Tuple[str, str, str, str, str, str]:
+    """
+    Main prediction function that handles the epitope prediction workflow
+    """
+    try:
+        # Input validation
+        if not pdb_file and not pdb_id:
+            return "Error: Please provide either a PDB ID or upload a PDB file", "", "", "", "", ""
+        
+        if pdb_id and not validate_pdb_id(pdb_id):
+            return "Error: PDB ID must be exactly 4 characters (letters and numbers)", "", "", "", "", ""
+        
+        if not validate_chain_id(chain_id):
+            return "Error: Chain ID must be exactly 1 character", "", "", "", "", ""
+        
+        # Update progress
+        if progress:
+            progress(0.1, desc="Initializing prediction...")
+        
+        # Process device configuration
+        device_id = -1 if device_config == "CPU Only" else int(device_config.split(" ")[1])
+        use_gpu = device_id >= 0
+        
+        # Load protein structure
+        if progress:
+            progress(0.2, desc="Loading protein structure...")
+        
+        antigen_chain = None
+        temp_file_path = None
+        
+        try:
+            if pdb_file:
+                # Handle uploaded file
+                if progress:
+                        progress(0.25, desc="Processing uploaded PDB file...")
+                
+                # Debug: print type and attributes of pdb_file
+                print(f"🔍 Debug: pdb_file type = {type(pdb_file)}")
+                print(f"🔍 Debug: pdb_file attributes = {dir(pdb_file)}")
+                
+                # Extract PDB ID from filename if not provided
+                if not pdb_id:
+                    if hasattr(pdb_file, 'name'):
+                        pdb_id = Path(pdb_file.name).stem.split('_')[0][:4]
+                    else:
+                        pdb_id = "UNKN"  # Default fallback
+                
+                # Save uploaded file to data/pdb/ directory with proper naming
+                timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+                filename = f"{pdb_id}_{chain_id}_{timestamp}.pdb"
+                temp_file_path = PDB_DATA_DIR / filename
+                
+                # Properly read and write the uploaded file
+                try:
+                    if hasattr(pdb_file, 'name') and os.path.isfile(pdb_file.name):
+                        # pdb_file is a file object with .name attribute
+                        print(f"📁 Processing file object: {pdb_file.name}")
+                        with open(pdb_file.name, "rb") as src:
+                            with open(temp_file_path, "wb") as dst:
+                                dst.write(src.read())
+                    elif hasattr(pdb_file, 'read'):
+                        # pdb_file is a file-like object
+                        print(f"📄 Processing file-like object")
+                        with open(temp_file_path, "wb") as f:
+                            f.write(pdb_file.read())
+                    else:
+                        # pdb_file is a string (file path)
+                        print(f"📍 Processing file path: {pdb_file}")
+                        with open(str(pdb_file), "rb") as src:
+                            with open(temp_file_path, "wb") as dst:
+                                dst.write(src.read())
+                    
+                    print(f"✅ PDB file saved to: {temp_file_path}")
+                    
+                except Exception as file_error:
+                    print(f"❌ Error processing uploaded file: {file_error}")
+                    return f"Error processing uploaded file: {str(file_error)}", "", "", "", "", ""
+                
+                antigen_chain = AntigenChain.from_pdb(
+                    path=str(temp_file_path),
+                    chain_id=chain_id,
+                    id=pdb_id
+                )
+            else:
+                # Load from PDB ID
+                if progress:
+                    progress(0.25, desc=f"Downloading PDB structure {pdb_id}...")
+                
+                antigen_chain = AntigenChain.from_pdb(
+                    chain_id=chain_id,
+                    id=pdb_id
+                )
+                
+        except Exception as e:
+            return f"Error loading protein structure: {str(e)}", "", "", "", "", ""
+        
+        if antigen_chain is None:
+            return "Error: Failed to load protein structure", "", "", "", "", ""
+        
+        # Run prediction
+        if progress:
+            progress(0.4, desc="Running epitope prediction...")
+        
+        try:
+            # Use threshold only if checkbox is checked
+            final_threshold = threshold if use_threshold else None
+            
+            predict_results = antigen_chain.predict(
+                model_path=DEFAULT_MODEL_PATH,
+                device_id=device_id,
+                radius=radius,
+                k=k,
+                threshold=final_threshold,
+                verbose=True,
+                encoder=encoder,
+                use_gpu=use_gpu,
+                auto_cleanup=auto_cleanup
+            )
+        except Exception as e:
+            error_msg = f"Error during prediction: {str(e)}"
+            print(f"Prediction error: {error_msg}")
+            import traceback
+            traceback.print_exc()
+            return error_msg, "", "", "", "", ""
+        
+        if progress:
+            progress(0.8, desc="Processing results...")
+        
+        # Process results
+        if not predict_results:
+            return "Error: No prediction results generated", "", "", "", "", ""
+        
+        # Extract prediction data
+        predicted_epitopes = predict_results.get("predicted_epitopes", [])
+        predictions = predict_results.get("predictions", {})
+        top_k_centers = predict_results.get("top_k_centers", [])
+        top_k_region_residues = predict_results.get("top_k_region_residues", [])
+        top_k_regions = predict_results.get("top_k_regions", [])
+        
+        # Calculate summary statistics
+        protein_length = len(antigen_chain.sequence)
+        epitope_count = len(predicted_epitopes)
+        region_count = len(top_k_regions)
+        coverage_rate = (len(top_k_region_residues) / protein_length) * 100 if protein_length > 0 else 0
+        
+        # Create summary text
+        summary_text = f"""
+## Prediction Results for {pdb_id}_{chain_id}
+
+### Protein Information
+- **PDB ID**: {pdb_id}
+- **Chain**: {chain_id}
+- **Length**: {protein_length} residues
+- **Sequence**: <div style="word-wrap: break-word; word-break: break-all; white-space: pre-wrap; max-width: 100%; font-family: monospace; background: #f5f5f5; padding: 8px; border-radius: 4px; margin: 5px 0; display: inline-block;">{antigen_chain.sequence}</div>
+
+### Prediction Summary
+- **Predicted Epitopes**: {epitope_count}
+- **Top-k Regions**: {region_count}
+- **Coverage Rate**: {coverage_rate:.1f}%
+
+### Top-k Region Centers
+{', '.join(map(str, top_k_centers))}
+
+### Predicted Epitope Residues
+{', '.join(map(str, predicted_epitopes))}
+
+### Binding Region Residues (Top-k Union)
+{', '.join(map(str, top_k_region_residues))}
+        """
+        
+        # Create epitope list text with residue names
+        epitope_text = f"Predicted Epitope Residues ({len(predicted_epitopes)}):\n"
+        epitope_lines = []
+        for res in predicted_epitopes:
+            # Get residue index from residue number
+            if res in antigen_chain.resnum_to_index:
+                res_idx = antigen_chain.resnum_to_index[res]
+                res_name = antigen_chain.sequence[res_idx]
+                epitope_lines.append(f"Residue {res} ({res_name})")
+            else:
+                epitope_lines.append(f"Residue {res}")
+        epitope_text += "\n".join(epitope_lines)
+        
+        # Create binding region text with residue names
+        binding_text = f"Binding Region Residues ({len(top_k_region_residues)}):\n"
+        binding_lines = []
+        for res in top_k_region_residues:
+            # Get residue index from residue number
+            if res in antigen_chain.resnum_to_index:
+                res_idx = antigen_chain.resnum_to_index[res]
+                res_name = antigen_chain.sequence[res_idx]
+                binding_lines.append(f"Residue {res} ({res_name})")
+            else:
+                binding_lines.append(f"Residue {res}")
+        binding_text += "\n".join(binding_lines)
+        
+        # Create downloadable files
+        if progress:
+            progress(0.9, desc="Preparing download files...")
+        
+        # JSON file
+        json_data = {
+            "protein_info": {
+                "id": pdb_id,
+                "chain_id": chain_id,
+                "length": protein_length,
+                "sequence": antigen_chain.sequence
+            },
+            "prediction": {
+                "predicted_epitopes": predicted_epitopes,
+                "predictions": predictions,
+                "top_k_centers": top_k_centers,
+                "top_k_region_residues": top_k_region_residues,
+                "top_k_regions": [
+                    {
+                        "center_idx": region.get('center_idx', 0),
+                        "graph_pred": region.get('graph_pred', 0),
+                        "covered_indices": region.get('covered_indices', [])
+                    }
+                    for region in top_k_regions
+                ],
+                "coverage_rate": coverage_rate,
+                "mean_region_value": 0 # No longer calculated
+            },
+            "parameters": {
+                "radius": radius,
+                "k": k,
+                "encoder": encoder,
+                "device_config": device_config,
+                "use_threshold": use_threshold,
+                "threshold": final_threshold,
+                "auto_cleanup": auto_cleanup
+            }
+        }
+        
+        # Save JSON file
+        json_file_path = tempfile.mktemp(suffix=".json")
+        with open(json_file_path, "w") as f:
+            json.dump(json_data, f, indent=2)
+        
+        # CSV file  
+        csv_data = []
+        for i, residue_num in enumerate(antigen_chain.residue_index):
+            residue_num = int(residue_num)
+            csv_data.append({
+                "Residue_Number": residue_num,
+                "Residue_Type": antigen_chain.sequence[i],
+                "Prediction_Probability": predictions.get(residue_num, 0.0),
+                "Is_Predicted_Epitope": 1 if residue_num in predicted_epitopes else 0,
+                "Is_In_TopK_Regions": 1 if residue_num in top_k_region_residues else 0
+            })
+        
+        csv_df = pd.DataFrame(csv_data)
+        csv_file_path = tempfile.mktemp(suffix=".csv")
+        csv_df.to_csv(csv_file_path, index=False)
+        
+        # Create 3D visualization
+        if progress:
+            progress(0.95, desc="Creating 3D visualization...")
+        
+        # Generate PDB string for visualization HTML file
+        html_file_path = None
+        try:
+            pdb_str = generate_pdb_string(antigen_chain)
+            html_content = create_pdb_visualization_html(
+                pdb_str, predicted_epitopes, predictions, f"{pdb_id}_{chain_id}", top_k_regions
+            )
+            
+            # Save HTML file to data directory for download
+            timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+            html_filename = f"{pdb_id}_{chain_id}_visualization_{timestamp}.html"
+            html_file_path = PDB_DATA_DIR / html_filename
+            
+            with open(html_file_path, "w", encoding='utf-8') as f:
+                f.write(html_content)
+                
+            print(f"✅ 3D visualization HTML saved to: {html_file_path}")
+                
+        except Exception as e:
+            html_file_path = None
+            print(f"Warning: Could not create 3D visualization: {str(e)}")
+        
+        # Clean up temporary files if auto_cleanup is enabled
+        if auto_cleanup and temp_file_path and os.path.exists(temp_file_path):
+            os.remove(temp_file_path)
+            print(f"🧹 Cleaned up temporary file: {temp_file_path}")
+        elif temp_file_path and os.path.exists(temp_file_path):
+            print(f"📁 PDB file retained at: {temp_file_path}")
+        
+        if progress:
+            progress(1.0, desc="Prediction completed!")
+        
+        # Return all results including HTML file path for download
+        return (
+            summary_text,
+            epitope_text,
+            binding_text,
+            str(html_file_path) if html_file_path else None,  # HTML file moved to 4th position
+            json_file_path,
+            csv_file_path
+        )
+        
+    except Exception as e:
+        import traceback
+        error_msg = f"Error: {str(e)}\n\nTraceback:\n{traceback.format_exc()}"
+        return error_msg, "", "", "", "", ""
+
+def generate_pdb_string(antigen_chain) -> str:
+    """Generate PDB string for 3D visualization"""
+    from esm.utils import residue_constants as RC
+    
+    pdb_str = "MODEL        1\n"
+    atom_num = 1
+    
+    for res_idx in range(len(antigen_chain.sequence)):
+        one_letter = antigen_chain.sequence[res_idx]
+        resname = antigen_chain.convert_letter_1to3(one_letter)
+        resnum = antigen_chain.residue_index[res_idx]
+        
+        mask = antigen_chain.atom37_mask[res_idx]
+        coords = antigen_chain.atom37_positions[res_idx][mask]
+        atoms = [name for name, exists in zip(RC.atom_types, mask) if exists]
+        
+        for atom_name, coord in zip(atoms, coords):
+            x, y, z = coord
+            pdb_str += (f"ATOM  {atom_num:5d}  {atom_name:<3s} {resname:>3s} {antigen_chain.chain_id:1s}{resnum:4d}"
+                       f"    {x:8.3f}{y:8.3f}{z:8.3f}  1.00  0.00\n")
+            atom_num += 1
+    
+    pdb_str += "ENDMDL\n"
+    return pdb_str
+
+def create_interface():
+    """Create the Gradio interface"""
+
+    with gr.Blocks(css="""
+        .container {
+            max-width: 1200px;
+            margin: 0 auto;
+            padding: 20px;
+        }
+        .header {
+            text-align: center;
+            margin-bottom: 30px;
+            padding: 20px;
+            background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+            color: white;
+            border-radius: 10px;
+        }
+        .header h1 {
+            font-size: 2.5em;
+            margin-bottom: 10px;
+        }
+        .form-row {
+            display: flex;
+            gap: 20px;
+            align-items: end;
+        }
+        .form-row > * {
+            flex: 1;
+        }
+        .section {
+            margin: 20px 0;
+            padding: 15px;
+            background: #f8f9fa;
+            border-radius: 8px;
+            border-left: 4px solid #007bff;
+        }
+        .section h2 {
+            color: #333;
+            margin-bottom: 15px;
+        }
+        .results-section {
+            margin-top: 30px;
+            padding: 20px;
+            background: #f0f8ff;
+            border-radius: 8px;
+            border: 1px solid #e0e8f0;
+        }
+        .download-section {
+            margin-top: 20px;
+            padding: 15px;
+            background: #f9f9f9;
+            border-radius: 8px;
+        }
+        .download-section h3 {
+            color: #333;
+            margin-bottom: 10px;
+        }
+        """) as interface:
+        
+        # Header
+        gr.HTML("""
+        <div class="header">
+            <h1>🧬 B-cell Epitope Prediction Server</h1>
+            <p>Predict epitopes using the ReCEP model</p>
+        </div>
+        """)
+
+        with gr.Row():
+            with gr.Column(scale=1):
+                gr.HTML("<div class='section'><h2>📋 Input Protein Structure</h2></div>")
+
+                input_method = gr.Radio(
+                    choices=["PDB ID", "Upload PDB File"],
+                    value="PDB ID",
+                    label="Input Method"
+                )
+
+                pdb_id = gr.Textbox(
+                    label="PDB ID", 
+                    placeholder="e.g., 5I9Q", 
+                    max_lines=1, 
+                    visible=True
+                )
+                pdb_file = gr.File(
+                    label="Upload PDB File", 
+                    file_types=[".pdb", ".ent"], 
+                    visible=False
+                )
+                chain_id = gr.Textbox(
+                    label="Chain ID", 
+                    value="A", 
+                    max_lines=1
+                )
+
+                with gr.Accordion("🔧 Advanced Parameters", open=False):
+                    radius = gr.Slider(
+                        label="Radius (Å)", 
+                        minimum=1.0, 
+                        maximum=50.0, 
+                        step=0.1, 
+                        value=19.0
+                    )
+                    k = gr.Slider(
+                        label="Top-k Regions", 
+                        minimum=1, 
+                        maximum=20, 
+                        step=1, 
+                        value=7
+                    )
+                    encoder = gr.Dropdown(
+                        label="Encoder", 
+                        choices=["esmc", "esm2"], 
+                        value="esmc"
+                    )
+                    device_config = gr.Dropdown(
+                        label="Device Configuration", 
+                        choices=["CPU Only", "GPU 0", "GPU 1", "GPU 2", "GPU 3"], 
+                        value="CPU Only"
+                    )
+                    use_threshold = gr.Checkbox(
+                        label="Use Custom Threshold", 
+                        value=False
+                    )
+                    threshold = gr.Number(
+                        label="Threshold Value", 
+                        value=0.366, 
+                        visible=False
+                    )
+                    auto_cleanup = gr.Checkbox(
+                        label="Auto-cleanup Generated Data", 
+                        value=True
+                    )
+
+                predict_btn = gr.Button("🧮 Predict Epitopes", variant="primary", size="lg")
+
+            with gr.Column(scale=2):
+                gr.HTML("<div class='section'><h2>📊 Prediction Results</h2></div>")
+
+                results_text = gr.Markdown(label="Prediction Summary", visible=True)
+
+                # 3D Visualization download (moved to top)
+                gr.HTML("<div style='margin: 15px 0; padding: 10px; background: #f0f8ff; border-left: 4px solid #4a90e2; border-radius: 5px;'><h3 style='margin: 0 0 8px 0; color: #333;'>🧬 3D Visualization</h3><p style='margin: 0; color: #666;'>You can download the HTML to visualize the prediction results and the spheres used.</p></div>")
+                html_download = gr.File(
+                    label="Download Interactive 3D Visualization HTML",
+                    visible=True
+                )
+
+                with gr.Row():
+                    epitope_list = gr.Textbox(
+                        label="Predicted Epitope Residues", 
+                        max_lines=10, 
+                        interactive=False
+                    )
+                    binding_regions = gr.Textbox(
+                        label="Binding Region Residues", 
+                        max_lines=10, 
+                        interactive=False
+                    )
+
+                gr.HTML("<div class='download-section'><h3>📥 Download Data Results</h3></div>")
+                with gr.Row():
+                    json_download = gr.File(
+                        label="JSON Results",
+                        visible=True
+                    )
+                    csv_download = gr.File(
+                        label="CSV Results",
+                        visible=True
+                    )
+
+        def toggle_input_method(method):
+            return (gr.update(visible=method == "PDB ID"),
+                    gr.update(visible=method == "Upload PDB File"))
+
+        def toggle_threshold(use_threshold):
+            return gr.update(visible=use_threshold)
+
+        input_method.change(toggle_input_method, inputs=[input_method], outputs=[pdb_id, pdb_file])
+        use_threshold.change(toggle_threshold, inputs=[use_threshold], outputs=[threshold])
+
+        predict_btn.click(
+            predict_epitopes,
+            inputs=[
+                pdb_id, pdb_file, chain_id, radius, k, encoder, 
+                device_config, use_threshold, threshold, auto_cleanup
+            ],
+            outputs=[
+                results_text, epitope_list, binding_regions, 
+                html_download, json_download, csv_download
+            ],
+            show_progress=True
+        )
+
+        gr.HTML("""
+        <div style="text-align: center; margin-top: 30px; padding: 20px; background: #f0f0f0; border-radius: 10px;">
+            <p>© 2024 B-cell Epitope Prediction Server | Powered by ReCEP model</p>
+            <p>🚀 Advanced AI-powered epitope prediction with interactive 3D visualization</p>
+            <p><strong>Features:</strong> PDB ID/File support • ESM-C encoder • GPU acceleration • 3D visualization • Multiple export formats</p>
+        </div>
+        """)
+
+    return interface
+
+
+if __name__ == "__main__":
+    # Create and launch the interface
+    try:
+        interface = create_interface()
+        
+        # Check if running on Hugging Face Spaces
+        is_spaces = os.getenv("SPACE_ID") is not None
+        
+        interface.launch(
+            server_name="0.0.0.0",
+            server_port=7860,
+            share=is_spaces,  # Use share=True on Spaces, False locally
+            show_error=True,
+            max_threads=4 if is_spaces else 8
+        )
+    except Exception as e:
+        print(f"Error launching application: {e}")
+        print("Please ensure all dependencies are installed correctly.")
+        import traceback
+        traceback.print_exc()

config.yaml

@@ -0,0 +1,16 @@
+title: B-cell Epitope Prediction Server
+emoji: 🧬
+colorFrom: blue
+colorTo: purple
+sdk: gradio
+sdk_version: 4.44.1
+app_file: app.py
+pinned: false
+license: mit
+short_description: "AI-powered B-cell epitope prediction using the ReCEP model"
+tags:
+  - bioinformatics
+  - protein-structure
+  - epitope-prediction
+  - machine-learning
+  - computational-biology 

data/epitopes/data_splits.json

@@ -0,0 +1,2878 @@
+{
+  "train": [
+    [
+      "6elu",
+      "A"
+    ],
+    [
+      "4zfg",
+      "A"
+    ],
+    [
+      "8tfl",
+      "B"
+    ],
+    [
+      "6gv4",
+      "B"
+    ],
+    [
+      "3lh2",
+      "V"
+    ],
+    [
+      "5b3j",
+      "C"
+    ],
+    [
+      "5bv7",
+      "A"
+    ],
+    [
+      "6u6u",
+      "R"
+    ],
+    [
+      "1fj1",
+      "E"
+    ],
+    [
+      "8tv3",
+      "A"
+    ],
+    [
+      "3vg9",
+      "A"
+    ],
+    [
+      "7kql",
+      "T"
+    ],
+    [
+      "1bgx",
+      "T"
+    ],
+    [
+      "9ima",
+      "A"
+    ],
+    [
+      "5fcu",
+      "G"
+    ],
+    [
+      "6v4n",
+      "W"
+    ],
+    [
+      "8u1c",
+      "A"
+    ],
+    [
+      "7tpd",
+      "A"
+    ],
+    [
+      "9ivk",
+      "A"
+    ],
+    [
+      "8zd5",
+      "D"
+    ],
+    [
+      "8r8d",
+      "B"
+    ],
+    [
+      "8yxi",
+      "A"
+    ],
+    [
+      "6y9b",
+      "C"
+    ],
+    [
+      "7tuf",
+      "C"
+    ],
+    [
+      "8gkl",
+      "E"
+    ],
+    [
+      "1lk3",
+      "B"
+    ],
+    [
+      "6wgl",
+      "C"
+    ],
+    [
+      "1egj",
+      "A"
+    ],
+    [
+      "6k65",
+      "A"
+    ],
+    [
+      "7lxx",
+      "A"
+    ],
+    [
+      "3kr3",
+      "D"
+    ],
+    [
+      "4f37",
+      "A"
+    ],
+    [
+      "6fgb",
+      "A"
+    ],
+    [
+      "5e8e",
+      "H"
+    ],
+    [
+      "6iw2",
+      "D"
+    ],
+    [
+      "7lf8",
+      "A"
+    ],
+    [
+      "1nsn",
+      "S"
+    ],
+    [
+      "4plj",
+      "A"
+    ],
+    [
+      "3rkd",
+      "A"
+    ],
+    [
+      "8y0r",
+      "2"
+    ],
+    [
+      "8y0q",
+      "2"
+    ],
+    [
+      "6hx4",
+      "A"
+    ],
+    [
+      "6wzk",
+      "E"
+    ],
+    [
+      "6wzj",
+      "E"
+    ],
+    [
+      "5ob5",
+      "A"
+    ],
+    [
+      "5hys",
+      "G"
+    ],
+    [
+      "7shu",
+      "A"
+    ],
+    [
+      "5veb",
+      "Y"
+    ],
+    [
+      "7zli",
+      "A"
+    ],
+    [
+      "6wbv",
+      "A"
+    ],
+    [
+      "4rdq",
+      "C"
+    ],
+    [
+      "8caf",
+      "E"
+    ],
+    [
+      "4hwb",
+      "A"
+    ],
+    [
+      "6ck9",
+      "G"
+    ],
+    [
+      "6mug",
+      "G"
+    ],
+    [
+      "6vzi",
+      "G"
+    ],
+    [
+      "4ma7",
+      "A"
+    ],
+    [
+      "4yxl",
+      "A"
+    ],
+    [
+      "1tqb",
+      "A"
+    ],
+    [
+      "4yxk",
+      "A"
+    ],
+    [
+      "6sv2",
+      "A"
+    ],
+    [
+      "8dyx",
+      "I"
+    ],
+    [
+      "7l7r",
+      "G"
+    ],
+    [
+      "6umx",
+      "A"
+    ],
+    [
+      "6wtu",
+      "A"
+    ],
+    [
+      "3l95",
+      "X"
+    ],
+    [
+      "8v2e",
+      "C"
+    ],
+    [
+      "4ala",
+      "C"
+    ],
+    [
+      "6wo4",
+      "E"
+    ],
+    [
+      "6uyf",
+      "E"
+    ],
+    [
+      "7jtg",
+      "E"
+    ],
+    [
+      "6bkc",
+      "E"
+    ],
+    [
+      "5ldn",
+      "A"
+    ],
+    [
+      "6nms",
+      "C"
+    ],
+    [
+      "4edw",
+      "V"
+    ],
+    [
+      "5vyf",
+      "F"
+    ],
+    [
+      "6qb6",
+      "A"
+    ],
+    [
+      "3cvh",
+      "M"
+    ],
+    [
+      "8tq8",
+      "C"
+    ],
+    [
+      "8tqa",
+      "A"
+    ],
+    [
+      "4xvu",
+      "H"
+    ],
+    [
+      "4xwo",
+      "A"
+    ],
+    [
+      "2wuc",
+      "A"
+    ],
+    [
+      "8jxs",
+      "A"
+    ],
+    [
+      "8hrx",
+      "B"
+    ],
+    [
+      "4ypg",
+      "D"
+    ],
+    [
+      "8urf",
+      "A"
+    ],
+    [
+      "8j8r",
+      "B"
+    ],
+    [
+      "4cni",
+      "D"
+    ],
+    [
+      "4o9h",
+      "A"
+    ],
+    [
+      "4zs7",
+      "A"
+    ],
+    [
+      "5ukr",
+      "G"
+    ],
+    [
+      "4a6y",
+      "A"
+    ],
+    [
+      "7kq7",
+      "B"
+    ],
+    [
+      "8wre",
+      "A"
+    ],
+    [
+      "3qwo",
+      "C"
+    ],
+    [
+      "3gi9",
+      "C"
+    ],
+    [
+      "8f5i",
+      "A"
+    ],
+    [
+      "3s35",
+      "X"
+    ],
+    [
+      "1ncc",
+      "N"
+    ],
+    [
+      "6lxj",
+      "D"
+    ],
+    [
+      "4kxz",
+      "D"
+    ],
+    [
+      "7v23",
+      "B"
+    ],
+    [
+      "9c44",
+      "C"
+    ],
+    [
+      "7k9j",
+      "C"
+    ],
+    [
+      "8hhy",
+      "A"
+    ],
+    [
+      "7e7x",
+      "A"
+    ],
+    [
+      "6ogx",
+      "G"
+    ],
+    [
+      "7yk4",
+      "B"
+    ],
+    [
+      "4dw2",
+      "U"
+    ],
+    [
+      "5k9k",
+      "F"
+    ],
+    [
+      "8ut3",
+      "C"
+    ],
+    [
+      "4o58",
+      "A"
+    ],
+    [
+      "6umg",
+      "C"
+    ],
+    [
+      "5ikc",
+      "N"
+    ],
+    [
+      "6tyb",
+      "G"
+    ],
+    [
+      "7dm2",
+      "A"
+    ],
+    [
+      "3v6o",
+      "B"
+    ],
+    [
+      "4u6v",
+      "B"
+    ],
+    [
+      "4khx",
+      "A"
+    ],
+    [
+      "5w2b",
+      "A"
+    ],
+    [
+      "6mlk",
+      "A"
+    ],
+    [
+      "4ywg",
+      "G"
+    ],
+    [
+      "8b7h",
+      "A"
+    ],
+    [
+      "7upw",
+      "A"
+    ],
+    [
+      "8r1d",
+      "C"
+    ],
+    [
+      "7q0i",
+      "D"
+    ],
+    [
+      "7ttx",
+      "A"
+    ],
+    [
+      "8y6a",
+      "B"
+    ],
+    [
+      "6xkq",
+      "A"
+    ],
+    [
+      "3o2d",
+      "A"
+    ],
+    [
+      "1xf5",
+      "L"
+    ],
+    [
+      "5u3d",
+      "E"
+    ],
+    [
+      "8d9y",
+      "K"
+    ],
+    [
+      "1iqd",
+      "C"
+    ],
+    [
+      "8c7m",
+      "B"
+    ],
+    [
+      "7mlh",
+      "F"
+    ],
+    [
+      "6oor",
+      "A"
+    ],
+    [
+      "3uc0",
+      "B"
+    ],
+    [
+      "2hfg",
+      "R"
+    ],
+    [
+      "8too",
+      "I"
+    ],
+    [
+      "7a0x",
+      "B"
+    ],
+    [
+      "4krp",
+      "A"
+    ],
+    [
+      "3ks0",
+      "A"
+    ],
+    [
+      "8vgm",
+      "A"
+    ],
+    [
+      "6o9i",
+      "C"
+    ],
+    [
+      "8y6i",
+      "A"
+    ],
+    [
+      "6xsw",
+      "X"
+    ],
+    [
+      "8tbq",
+      "P"
+    ],
+    [
+      "6mvl",
+      "A"
+    ],
+    [
+      "5epm",
+      "C"
+    ],
+    [
+      "6ppg",
+      "G"
+    ],
+    [
+      "8ruu",
+      "Y"
+    ],
+    [
+      "6wmw",
+      "B"
+    ],
+    [
+      "6svl",
+      "J"
+    ],
+    [
+      "6vtw",
+      "A"
+    ],
+    [
+      "3b9k",
+      "F"
+    ],
+    [
+      "7nx3",
+      "F"
+    ],
+    [
+      "5hbv",
+      "B"
+    ],
+    [
+      "3grw",
+      "A"
+    ],
+    [
+      "4mwf",
+      "D"
+    ],
+    [
+      "6meh",
+      "C"
+    ],
+    [
+      "6uyd",
+      "F"
+    ],
+    [
+      "6wo5",
+      "E"
+    ],
+    [
+      "1xiw",
+      "B"
+    ],
+    [
+      "7bbj",
+      "B"
+    ],
+    [
+      "6xcj",
+      "G"
+    ],
+    [
+      "8jlx",
+      "A"
+    ],
+    [
+      "8jlw",
+      "A"
+    ],
+    [
+      "7s7i",
+      "B"
+    ],
+    [
+      "6ddm",
+      "C"
+    ],
+    [
+      "4yqx",
+      "M"
+    ],
+    [
+      "4yue",
+      "C"
+    ],
+    [
+      "8dcn",
+      "F"
+    ],
+    [
+      "4i3r",
+      "G"
+    ],
+    [
+      "7n3c",
+      "C"
+    ],
+    [
+      "7sue",
+      "D"
+    ],
+    [
+      "6aod",
+      "C"
+    ],
+    [
+      "4aei",
+      "C"
+    ],
+    [
+      "7cj2",
+      "A"
+    ],
+    [
+      "5xez",
+      "A"
+    ],
+    [
+      "5yoy",
+      "A"
+    ],
+    [
+      "5wux",
+      "E"
+    ],
+    [
+      "5mo9",
+      "X"
+    ],
+    [
+      "4np4",
+      "A"
+    ],
+    [
+      "6pi7",
+      "A"
+    ],
+    [
+      "7usl",
+      "C"
+    ],
+    [
+      "7rah",
+      "E"
+    ],
+    [
+      "1ob1",
+      "F"
+    ],
+    [
+      "8dfh",
+      "A"
+    ],
+    [
+      "6xqw",
+      "E"
+    ],
+    [
+      "4f2m",
+      "E"
+    ],
+    [
+      "2j88",
+      "A"
+    ],
+    [
+      "4ps4",
+      "A"
+    ],
+    [
+      "8slb",
+      "A"
+    ],
+    [
+      "8jkf",
+      "A"
+    ],
+    [
+      "4xp9",
+      "C"
+    ],
+    [
+      "1uwx",
+      "B"
+    ],
+    [
+      "6wfw",
+      "G"
+    ],
+    [
+      "6u8c",
+      "B"
+    ],
+    [
+      "4lmq",
+      "F"
+    ],
+    [
+      "8w86",
+      "D"
+    ],
+    [
+      "8w83",
+      "L"
+    ],
+    [
+      "8w85",
+      "H"
+    ],
+    [
+      "6xxv",
+      "C"
+    ],
+    [
+      "9dez",
+      "C"
+    ],
+    [
+      "5w5z",
+      "A"
+    ],
+    [
+      "1cl7",
+      "I"
+    ],
+    [
+      "8tco",
+      "A"
+    ],
+    [
+      "6c9u",
+      "A"
+    ],
+    [
+      "7ox4",
+      "C"
+    ],
+    [
+      "6a77",
+      "A"
+    ],
+    [
+      "6cw2",
+      "C"
+    ],
+    [
+      "8dn7",
+      "C"
+    ],
+    [
+      "5k59",
+      "B"
+    ],
+    [
+      "7xy8",
+      "A"
+    ],
+    [
+      "7daa",
+      "A"
+    ],
+    [
+      "3wih",
+      "A"
+    ],
+    [
+      "8tp7",
+      "C"
+    ],
+    [
+      "6iek",
+      "D"
+    ],
+    [
+      "7ul3",
+      "A"
+    ],
+    [
+      "5x2n",
+      "C"
+    ],
+    [
+      "7txt",
+      "S"
+    ],
+    [
+      "8e8x",
+      "1"
+    ],
+    [
+      "8e8y",
+      "1"
+    ],
+    [
+      "3d85",
+      "C"
+    ],
+    [
+      "6ztr",
+      "I"
+    ],
+    [
+      "8d9z",
+      "D"
+    ],
+    [
+      "3nh7",
+      "C"
+    ],
+    [
+      "8y6h",
+      "A"
+    ],
+    [
+      "8y31",
+      "E"
+    ],
+    [
+      "3hi6",
+      "A"
+    ],
+    [
+      "5jq6",
+      "A"
+    ],
+    [
+      "6ks0",
+      "A"
+    ],
+    [
+      "5f3b",
+      "C"
+    ],
+    [
+      "5th9",
+      "B"
+    ],
+    [
+      "5kw9",
+      "A"
+    ],
+    [
+      "6z3q",
+      "A"
+    ],
+    [
+      "5wi9",
+      "A"
+    ],
+    [
+      "6uj9",
+      "A"
+    ],
+    [
+      "3hae",
+      "J"
+    ],
+    [
+      "7t0l",
+      "D"
+    ],
+    [
+      "6id4",
+      "E"
+    ],
+    [
+      "3l5x",
+      "A"
+    ],
+    [
+      "5l6y",
+      "C"
+    ],
+    [
+      "7rew",
+      "I"
+    ],
+    [
+      "3l5w",
+      "J"
+    ],
+    [
+      "6yio",
+      "B"
+    ],
+    [
+      "3nfp",
+      "I"
+    ],
+    [
+      "2hmi",
+      "B"
+    ],
+    [
+      "8da0",
+      "E"
+    ],
+    [
+      "6wh9",
+      "D"
+    ],
+    [
+      "7rt9",
+      "Y"
+    ],
+    [
+      "7ec5",
+      "C"
+    ],
+    [
+      "6vvu",
+      "B"
+    ],
+    [
+      "6x3x",
+      "D"
+    ],
+    [
+      "7zoz",
+      "A"
+    ],
+    [
+      "5tl5",
+      "A"
+    ],
+    [
+      "5gzn",
+      "E"
+    ],
+    [
+      "7vux",
+      "A"
+    ],
+    [
+      "6jbt",
+      "F"
+    ],
+    [
+      "5wt9",
+      "G"
+    ],
+    [
+      "6j15",
+      "D"
+    ],
+    [
+      "5jxe",
+      "B"
+    ],
+    [
+      "6j14",
+      "G"
+    ],
+    [
+      "1rjl",
+      "C"
+    ],
+    [
+      "8uky",
+      "D"
+    ],
+    [
+      "7wg3",
+      "K"
+    ],
+    [
+      "7x28",
+      "F"
+    ],
+    [
+      "4ffv",
+      "B"
+    ],
+    [
+      "2oz4",
+      "A"
+    ],
+    [
+      "4hc1",
+      "A"
+    ],
+    [
+      "7lsg",
+      "C"
+    ],
+    [
+      "6j5d",
+      "A"
+    ],
+    [
+      "7kyl",
+      "Z"
+    ],
+    [
+      "7lse",
+      "E"
+    ],
+    [
+      "6cmi",
+      "B"
+    ],
+    [
+      "8rp8",
+      "D"
+    ],
+    [
+      "7wn8",
+      "C"
+    ],
+    [
+      "8jel",
+      "J"
+    ],
+    [
+      "4jlr",
+      "C"
+    ],
+    [
+      "1kb5",
+      "B"
+    ],
+    [
+      "8c7h",
+      "D"
+    ],
+    [
+      "2vxt",
+      "I"
+    ],
+    [
+      "4qhu",
+      "C"
+    ],
+    [
+      "5n7w",
+      "X"
+    ],
+    [
+      "7z2m",
+      "G"
+    ],
+    [
+      "6wio",
+      "C"
+    ],
+    [
+      "2vxs",
+      "A"
+    ],
+    [
+      "4nnp",
+      "B"
+    ],
+    [
+      "8f5n",
+      "A"
+    ],
+    [
+      "8iuy",
+      "A"
+    ],
+    [
+      "6adb",
+      "A"
+    ],
+    [
+      "5ggv",
+      "Y"
+    ],
+    [
+      "3ld8",
+      "A"
+    ],
+    [
+      "2q8a",
+      "A"
+    ],
+    [
+      "2q8b",
+      "A"
+    ],
+    [
+      "7n8q",
+      "A"
+    ],
+    [
+      "7ued",
+      "M"
+    ],
+    [
+      "7vgr",
+      "A"
+    ],
+    [
+      "7phu",
+      "A"
+    ],
+    [
+      "2r56",
+      "A"
+    ],
+    [
+      "7bsc",
+      "A"
+    ],
+    [
+      "6ewb",
+      "C"
+    ],
+    [
+      "6n81",
+      "A"
+    ],
+    [
+      "3wfd",
+      "B"
+    ],
+    [
+      "4i2x",
+      "E"
+    ],
+    [
+      "6mei",
+      "C"
+    ],
+    [
+      "7dc8",
+      "C"
+    ],
+    [
+      "8ts0",
+      "A"
+    ],
+    [
+      "8d7e",
+      "C"
+    ],
+    [
+      "5y9j",
+      "A"
+    ],
+    [
+      "4qci",
+      "D"
+    ],
+    [
+      "8sgi",
+      "A"
+    ],
+    [
+      "9gwt",
+      "P"
+    ],
+    [
+      "8oxv",
+      "A"
+    ],
+    [
+      "4lvo",
+      "A"
+    ],
+    [
+      "4d9q",
+      "A"
+    ],
+    [
+      "8ee0",
+      "A"
+    ],
+    [
+      "8aci",
+      "A"
+    ],
+    [
+      "5xj4",
+      "A"
+    ],
+    [
+      "5xxy",
+      "A"
+    ],
+    [
+      "7ket",
+      "C"
+    ],
+    [
+      "2ypv",
+      "A"
+    ],
+    [
+      "6al5",
+      "A"
+    ],
+    [
+      "2arj",
+      "R"
+    ],
+    [
+      "6m3b",
+      "A"
+    ],
+    [
+      "5e94",
+      "G"
+    ],
+    [
+      "4fqj",
+      "A"
+    ],
+    [
+      "6lyn",
+      "C"
+    ],
+    [
+      "4d3c",
+      "A"
+    ],
+    [
+      "3jwd",
+      "A"
+    ],
+    [
+      "7jkt",
+      "G"
+    ],
+    [
+      "4rwy",
+      "A"
+    ],
+    [
+      "4ye4",
+      "G"
+    ],
+    [
+      "3idx",
+      "G"
+    ],
+    [
+      "1g9n",
+      "G"
+    ],
+    [
+      "6vep",
+      "Q"
+    ],
+    [
+      "7kd6",
+      "W"
+    ],
+    [
+      "2h9g",
+      "R"
+    ],
+    [
+      "8sic",
+      "G"
+    ],
+    [
+      "5vpl",
+      "A"
+    ],
+    [
+      "5vcn",
+      "A"
+    ],
+    [
+      "2adf",
+      "A"
+    ],
+    [
+      "8ahn",
+      "A"
+    ],
+    [
+      "7ox3",
+      "C"
+    ],
+    [
+      "7ox1",
+      "X"
+    ],
+    [
+      "7lr3",
+      "C"
+    ],
+    [
+      "4qti",
+      "U"
+    ],
+    [
+      "7x8q",
+      "D"
+    ],
+    [
+      "2vxq",
+      "A"
+    ],
+    [
+      "1v7m",
+      "V"
+    ],
+    [
+      "5mhr",
+      "D"
+    ],
+    [
+      "8djg",
+      "F"
+    ],
+    [
+      "5occ",
+      "A"
+    ],
+    [
+      "4rgo",
+      "S"
+    ],
+    [
+      "3u9p",
+      "C"
+    ],
+    [
+      "4oii",
+      "B"
+    ],
+    [
+      "7df1",
+      "D"
+    ],
+    [
+      "8ath",
+      "A"
+    ],
+    [
+      "8pg0",
+      "A"
+    ],
+    [
+      "8kdm",
+      "B"
+    ],
+    [
+      "7ttm",
+      "A"
+    ],
+    [
+      "7rp2",
+      "A"
+    ],
+    [
+      "2uzi",
+      "R"
+    ],
+    [
+      "8g8d",
+      "P"
+    ],
+    [
+      "4g6f",
+      "F"
+    ],
+    [
+      "5u3m",
+      "A"
+    ],
+    [
+      "6p9h",
+      "A"
+    ],
+    [
+      "5eu7",
+      "B"
+    ],
+    [
+      "2qqk",
+      "A"
+    ],
+    [
+      "4tsa",
+      "A"
+    ],
+    [
+      "6u9s",
+      "C"
+    ],
+    [
+      "6hig",
+      "B"
+    ],
+    [
+      "6lxi",
+      "B"
+    ],
+    [
+      "4qnp",
+      "A"
+    ],
+    [
+      "6kyz",
+      "A"
+    ],
+    [
+      "6kz0",
+      "J"
+    ],
+    [
+      "2xqb",
+      "A"
+    ],
+    [
+      "5w3l",
+      "B"
+    ],
+    [
+      "7uvf",
+      "B"
+    ],
+    [
+      "3lev",
+      "A"
+    ],
+    [
+      "6v4p",
+      "B"
+    ],
+    [
+      "7rxl",
+      "F"
+    ],
+    [
+      "6sni",
+      "X"
+    ],
+    [
+      "7qu1",
+      "C"
+    ],
+    [
+      "3ehb",
+      "B"
+    ],
+    [
+      "6u2f",
+      "A"
+    ],
+    [
+      "6u36",
+      "B"
+    ],
+    [
+      "1dee",
+      "H"
+    ],
+    [
+      "1oaz",
+      "A"
+    ],
+    [
+      "8byu",
+      "A"
+    ],
+    [
+      "3u30",
+      "A"
+    ],
+    [
+      "3ru8",
+      "X"
+    ],
+    [
+      "6wzl",
+      "E"
+    ],
+    [
+      "9dwe",
+      "A"
+    ],
+    [
+      "3gbm",
+      "A"
+    ],
+    [
+      "5dur",
+      "A"
+    ],
+    [
+      "6iuv",
+      "A"
+    ],
+    [
+      "4xnm",
+      "C"
+    ],
+    [
+      "8txp",
+      "B"
+    ],
+    [
+      "7l0l",
+      "E"
+    ],
+    [
+      "4k3j",
+      "B"
+    ],
+    [
+      "7a3o",
+      "A"
+    ],
+    [
+      "6dfj",
+      "E"
+    ],
+    [
+      "5vic",
+      "E"
+    ],
+    [
+      "4jzj",
+      "D"
+    ],
+    [
+      "7mrz",
+      "C"
+    ],
+    [
+      "3mj9",
+      "A"
+    ],
+    [
+      "3q3g",
+      "I"
+    ],
+    [
+      "7uot",
+      "C"
+    ],
+    [
+      "7tyv",
+      "B"
+    ],
+    [
+      "8dg9",
+      "C"
+    ],
+    [
+      "6apb",
+      "C"
+    ],
+    [
+      "8ulj",
+      "B"
+    ],
+    [
+      "7chz",
+      "I"
+    ],
+    [
+      "7c61",
+      "A"
+    ],
+    [
+      "8fdo",
+      "C"
+    ],
+    [
+      "7sgm",
+      "B"
+    ],
+    [
+      "1nl0",
+      "G"
+    ],
+    [
+      "4j6r",
+      "G"
+    ],
+    [
+      "4xmp",
+      "G"
+    ],
+    [
+      "4lss",
+      "G"
+    ],
+    [
+      "1e6j",
+      "P"
+    ],
+    [
+      "6wit",
+      "C"
+    ],
+    [
+      "6was",
+      "G"
+    ],
+    [
+      "6ks1",
+      "A"
+    ],
+    [
+      "5fb8",
+      "C"
+    ],
+    [
+      "8u2c",
+      "A"
+    ],
+    [
+      "6pe8",
+      "U"
+    ],
+    [
+      "8yx9",
+      "J"
+    ],
+    [
+      "6blh",
+      "G"
+    ],
+    [
+      "6uvo",
+      "D"
+    ],
+    [
+      "5if0",
+      "I"
+    ],
+    [
+      "4jpk",
+      "A"
+    ],
+    [
+      "2jel",
+      "P"
+    ],
+    [
+      "8sak",
+      "B"
+    ],
+    [
+      "3u4e",
+      "J"
+    ],
+    [
+      "3u2s",
+      "G"
+    ],
+    [
+      "7n0u",
+      "C"
+    ],
+    [
+      "8y3u",
+      "K"
+    ],
+    [
+      "8v5q",
+      "G"
+    ],
+    [
+      "6ion",
+      "A"
+    ],
+    [
+      "8h3n",
+      "C"
+    ],
+    [
+      "7yms",
+      "A"
+    ],
+    [
+      "8db4",
+      "E"
+    ],
+    [
+      "6vgr",
+      "A"
+    ],
+    [
+      "8dao",
+      "F"
+    ],
+    [
+      "8g4t",
+      "I"
+    ],
+    [
+      "6nnf",
+      "G"
+    ],
+    [
+      "7kyo",
+      "B"
+    ],
+    [
+      "6xlq",
+      "A"
+    ],
+    [
+      "1nfd",
+      "D"
+    ],
+    [
+      "4ht1",
+      "T"
+    ],
+    [
+      "7u5b",
+      "J"
+    ],
+    [
+      "6ba5",
+      "N"
+    ],
+    [
+      "4okv",
+      "F"
+    ],
+    [
+      "5wk3",
+      "C"
+    ],
+    [
+      "7a3q",
+      "A"
+    ],
+    [
+      "4bz2",
+      "A"
+    ],
+    [
+      "3mxw",
+      "A"
+    ],
+    [
+      "1ors",
+      "C"
+    ],
+    [
+      "3q1s",
+      "I"
+    ],
+    [
+      "3wkm",
+      "B"
+    ],
+    [
+      "3lhp",
+      "S"
+    ],
+    [
+      "6s3d",
+      "M"
+    ],
+    [
+      "9dx6",
+      "A"
+    ],
+    [
+      "2j4w",
+      "D"
+    ],
+    [
+      "6z7z",
+      "F"
+    ],
+    [
+      "3w9e",
+      "C"
+    ],
+    [
+      "4dkf",
+      "A"
+    ],
+    [
+      "7ce2",
+      "A"
+    ],
+    [
+      "7upb",
+      "D"
+    ],
+    [
+      "7ki6",
+      "A"
+    ],
+    [
+      "5l0q",
+      "A"
+    ],
+    [
+      "7xw6",
+      "R"
+    ],
+    [
+      "6lz9",
+      "B"
+    ],
+    [
+      "7m3n",
+      "A"
+    ],
+    [
+      "7so5",
+      "A"
+    ],
+    [
+      "3tje",
+      "F"
+    ],
+    [
+      "3thm",
+      "F"
+    ],
+    [
+      "5te4",
+      "G"
+    ],
+    [
+      "6bgt",
+      "C"
+    ],
+    [
+      "4ag4",
+      "A"
+    ],
+    [
+      "7joo",
+      "C"
+    ],
+    [
+      "8tui",
+      "A"
+    ],
+    [
+      "3efd",
+      "K"
+    ],
+    [
+      "7r58",
+      "A"
+    ],
+    [
+      "2nyy",
+      "A"
+    ],
+    [
+      "8jnk",
+      "C"
+    ],
+    [
+      "8dn6",
+      "A"
+    ],
+    [
+      "4jr9",
+      "A"
+    ],
+    [
+      "6otc",
+      "A"
+    ],
+    [
+      "8tlm",
+      "C"
+    ],
+    [
+      "4uuj",
+      "C"
+    ],
+    [
+      "6by3",
+      "C"
+    ],
+    [
+      "5tud",
+      "A"
+    ],
+    [
+      "7xrz",
+      "Y"
+    ],
+    [
+      "4lu5",
+      "A"
+    ],
+    [
+      "4m1g",
+      "B"
+    ],
+    [
+      "9dh2",
+      "T"
+    ],
+    [
+      "6k7o",
+      "P"
+    ],
+    [
+      "7wvg",
+      "B"
+    ],
+    [
+      "8f38",
+      "C"
+    ],
+    [
+      "6ml8",
+      "A"
+    ],
+    [
+      "6q0l",
+      "A"
+    ],
+    [
+      "3gbn",
+      "B"
+    ],
+    [
+      "6cxy",
+      "C"
+    ],
+    [
+      "2aep",
+      "A"
+    ],
+    [
+      "8g3q",
+      "G"
+    ],
+    [
+      "6q20",
+      "A"
+    ],
+    [
+      "6n6b",
+      "A"
+    ],
+    [
+      "8gat",
+      "A"
+    ],
+    [
+      "5utz",
+      "A"
+    ],
+    [
+      "4zso",
+      "E"
+    ],
+    [
+      "6dkj",
+      "D"
+    ],
+    [
+      "7s13",
+      "C"
+    ],
+    [
+      "7x29",
+      "C"
+    ],
+    [
+      "8ol9",
+      "H"
+    ],
+    [
+      "8sxp",
+      "C"
+    ],
+    [
+      "8vvk",
+      "B"
+    ],
+    [
+      "8tv1",
+      "C"
+    ],
+    [
+      "4liq",
+      "E"
+    ],
+    [
+      "7lkf",
+      "A"
+    ],
+    [
+      "7zxk",
+      "C"
+    ],
+    [
+      "1jrh",
+      "I"
+    ],
+    [
+      "6osv",
+      "K"
+    ],
+    [
+      "9jbq",
+      "C"
+    ],
+    [
+      "6cyf",
+      "Q"
+    ],
+    [
+      "8vvm",
+      "I"
+    ],
+    [
+      "3skj",
+      "F"
+    ],
+    [
+      "2xqy",
+      "E"
+    ],
+    [
+      "7qu2",
+      "C"
+    ],
+    [
+      "8v52",
+      "B"
+    ],
+    [
+      "7wtf",
+      "D"
+    ],
+    [
+      "7xco",
+      "C"
+    ],
+    [
+      "9fjk",
+      "B"
+    ],
+    [
+      "8xi6",
+      "C"
+    ],
+    [
+      "7e72",
+      "E"
+    ],
+    [
+      "4lst",
+      "G"
+    ],
+    [
+      "5te7",
+      "G"
+    ],
+    [
+      "2ybr",
+      "I"
+    ],
+    [
+      "5ush",
+      "X"
+    ],
+    [
+      "5dhv",
+      "M"
+    ],
+    [
+      "8e8r",
+      "2"
+    ],
+    [
+      "8e8s",
+      "2"
+    ],
+    [
+      "5kn5",
+      "C"
+    ],
+    [
+      "6iap",
+      "A"
+    ],
+    [
+      "7amp",
+      "B"
+    ],
+    [
+      "8tfn",
+      "B"
+    ],
+    [
+      "3pnw",
+      "R"
+    ],
+    [
+      "6o39",
+      "C"
+    ],
+    [
+      "8x0t",
+      "A"
+    ],
+    [
+      "8da1",
+      "I"
+    ],
+    [
+      "6phb",
+      "I"
+    ],
+    [
+      "6phc",
+      "I"
+    ],
+    [
+      "6wix",
+      "G"
+    ],
+    [
+      "8k3c",
+      "B"
+    ],
+    [
+      "8ffe",
+      "A"
+    ],
+    [
+      "5otj",
+      "D"
+    ],
+    [
+      "5lsp",
+      "P"
+    ],
+    [
+      "6s5a",
+      "A"
+    ],
+    [
+      "3vi4",
+      "D"
+    ],
+    [
+      "5nh3",
+      "B"
+    ],
+    [
+      "3i50",
+      "E"
+    ],
+    [
+      "4wfg",
+      "B"
+    ],
+    [
+      "6pis",
+      "B"
+    ],
+    [
+      "6s8j",
+      "E"
+    ],
+    [
+      "3nps",
+      "A"
+    ],
+    [
+      "1kyo",
+      "P"
+    ],
+    [
+      "7quh",
+      "A"
+    ],
+    [
+      "8cz5",
+      "A"
+    ],
+    [
+      "8t03",
+      "B"
+    ],
+    [
+      "4k94",
+      "C"
+    ]
+  ],
+  "test": [
+    [
+      "6ktr",
+      "C"
+    ],
+    [
+      "7o9s",
+      "A"
+    ],
+    [
+      "2bdn",
+      "A"
+    ],
+    [
+      "8d1t",
+      "A"
+    ],
+    [
+      "9b2w",
+      "D"
+    ],
+    [
+      "6mej",
+      "C"
+    ],
+    [
+      "8dk6",
+      "E"
+    ],
+    [
+      "8vgn",
+      "I"
+    ],
+    [
+      "8f60",
+      "C"
+    ],
+    [
+      "8udz",
+      "A"
+    ],
+    [
+      "7kpj",
+      "E"
+    ],
+    [
+      "7o52",
+      "U"
+    ],
+    [
+      "5b71",
+      "E"
+    ],
+    [
+      "4zff",
+      "D"
+    ],
+    [
+      "4uu9",
+      "D"
+    ],
+    [
+      "7lr3",
+      "D"
+    ],
+    [
+      "7lr4",
+      "D"
+    ],
+    [
+      "1mhp",
+      "B"
+    ],
+    [
+      "1ztx",
+      "E"
+    ],
+    [
+      "6i8s",
+      "D"
+    ],
+    [
+      "7ujd",
+      "A"
+    ],
+    [
+      "3gbn",
+      "A"
+    ],
+    [
+      "8txp",
+      "A"
+    ],
+    [
+      "6uyn",
+      "A"
+    ],
+    [
+      "4cad",
+      "I"
+    ],
+    [
+      "1yjd",
+      "C"
+    ],
+    [
+      "5u8r",
+      "A"
+    ],
+    [
+      "5d72",
+      "B"
+    ],
+    [
+      "6bfq",
+      "G"
+    ],
+    [
+      "8a1e",
+      "A"
+    ],
+    [
+      "4irz",
+      "A"
+    ],
+    [
+      "6o3b",
+      "C"
+    ],
+    [
+      "7x8p",
+      "D"
+    ],
+    [
+      "2yss",
+      "C"
+    ],
+    [
+      "1uac",
+      "Y"
+    ],
+    [
+      "1fbi",
+      "Y"
+    ],
+    [
+      "5vjo",
+      "F"
+    ],
+    [
+      "1jhl",
+      "A"
+    ],
+    [
+      "5o1r",
+      "A"
+    ],
+    [
+      "7yru",
+      "A"
+    ],
+    [
+      "2qqn",
+      "A"
+    ],
+    [
+      "5w4l",
+      "G"
+    ],
+    [
+      "6bf4",
+      "A"
+    ],
+    [
+      "5h35",
+      "C"
+    ],
+    [
+      "5cbe",
+      "F"
+    ],
+    [
+      "5w5x",
+      "A"
+    ],
+    [
+      "7lfa",
+      "C"
+    ],
+    [
+      "7lf7",
+      "M"
+    ],
+    [
+      "8cdd",
+      "B"
+    ],
+    [
+      "7phw",
+      "D"
+    ],
+    [
+      "6hga",
+      "B"
+    ],
+    [
+      "6h3t",
+      "B"
+    ],
+    [
+      "5lcv",
+      "A"
+    ],
+    [
+      "5i9q",
+      "A"
+    ],
+    [
+      "6rlo",
+      "J"
+    ],
+    [
+      "2xra",
+      "A"
+    ],
+    [
+      "4nzr",
+      "M"
+    ],
+    [
+      "1sy6",
+      "A"
+    ],
+    [
+      "1xiw",
+      "A"
+    ],
+    [
+      "8vsj",
+      "A"
+    ],
+    [
+      "6m58",
+      "A"
+    ],
+    [
+      "7q6c",
+      "A"
+    ],
+    [
+      "8u03",
+      "C"
+    ],
+    [
+      "4wv1",
+      "F"
+    ],
+    [
+      "5bo1",
+      "A"
+    ],
+    [
+      "3liz",
+      "A"
+    ],
+    [
+      "7uij",
+      "D"
+    ],
+    [
+      "5tq0",
+      "A"
+    ],
+    [
+      "6uym",
+      "E"
+    ],
+    [
+      "4g7v",
+      "S"
+    ],
+    [
+      "8j80",
+      "A"
+    ],
+    [
+      "7unb",
+      "R"
+    ],
+    [
+      "6e63",
+      "A"
+    ],
+    [
+      "5vkd",
+      "A"
+    ],
+    [
+      "5nmv",
+      "K"
+    ],
+    [
+      "6wzm",
+      "F"
+    ],
+    [
+      "7tuy",
+      "R"
+    ],
+    [
+      "7ahu",
+      "D"
+    ],
+    [
+      "7zqt",
+      "E"
+    ],
+    [
+      "8w84",
+      "C"
+    ],
+    [
+      "8vyn",
+      "C"
+    ],
+    [
+      "4ot1",
+      "A"
+    ],
+    [
+      "7sem",
+      "F"
+    ],
+    [
+      "8t9z",
+      "A"
+    ],
+    [
+      "6rps",
+      "A"
+    ],
+    [
+      "6gku",
+      "A"
+    ],
+    [
+      "7uvi",
+      "C"
+    ],
+    [
+      "7uvs",
+      "C"
+    ],
+    [
+      "8vzn",
+      "A"
+    ],
+    [
+      "5d8j",
+      "A"
+    ],
+    [
+      "3t2n",
+      "A"
+    ],
+    [
+      "8tq9",
+      "B"
+    ],
+    [
+      "4qww",
+      "B"
+    ],
+    [
+      "1uj3",
+      "C"
+    ],
+    [
+      "7xnf",
+      "A"
+    ],
+    [
+      "6mi2",
+      "C"
+    ],
+    [
+      "6a3w",
+      "L"
+    ],
+    [
+      "3zkn",
+      "B"
+    ],
+    [
+      "7ly0",
+      "A"
+    ],
+    [
+      "7lxw",
+      "A"
+    ],
+    [
+      "4uta",
+      "B"
+    ],
+    [
+      "6flb",
+      "G"
+    ],
+    [
+      "2j5l",
+      "A"
+    ],
+    [
+      "6mto",
+      "T"
+    ],
+    [
+      "8tzw",
+      "C"
+    ],
+    [
+      "4m62",
+      "T"
+    ],
+    [
+      "6cw2",
+      "D"
+    ],
+    [
+      "6cw3",
+      "F"
+    ],
+    [
+      "3hmx",
+      "A"
+    ],
+    [
+      "4dtg",
+      "K"
+    ],
+    [
+      "4rrp",
+      "P"
+    ],
+    [
+      "5eii",
+      "I"
+    ],
+    [
+      "8qya",
+      "A"
+    ],
+    [
+      "8a44",
+      "A"
+    ],
+    [
+      "7zjl",
+      "A"
+    ],
+    [
+      "8a99",
+      "C"
+    ],
+    [
+      "7mjk",
+      "C"
+    ],
+    [
+      "7q6e",
+      "A"
+    ],
+    [
+      "8h07",
+      "B"
+    ],
+    [
+      "6p67",
+      "G"
+    ],
+    [
+      "1wej",
+      "F"
+    ],
+    [
+      "4ogy",
+      "A"
+    ],
+    [
+      "8fgx",
+      "C"
+    ],
+    [
+      "1kb5",
+      "A"
+    ],
+    [
+      "4f15",
+      "J"
+    ],
+    [
+      "4leo",
+      "C"
+    ],
+    [
+      "8vdl",
+      "C"
+    ],
+    [
+      "6h2y",
+      "D"
+    ],
+    [
+      "8bk2",
+      "B"
+    ],
+    [
+      "5d93",
+      "A"
+    ],
+    [
+      "4ij3",
+      "A"
+    ],
+    [
+      "3r1g",
+      "B"
+    ],
+    [
+      "6vn1",
+      "C"
+    ],
+    [
+      "6qig",
+      "A"
+    ],
+    [
+      "7k7h",
+      "A"
+    ],
+    [
+      "8hpk",
+      "A"
+    ],
+    [
+      "5e8d",
+      "A"
+    ],
+    [
+      "5bk2",
+      "A"
+    ],
+    [
+      "6nyq",
+      "C"
+    ],
+    [
+      "2zch",
+      "P"
+    ],
+    [
+      "8v91",
+      "B"
+    ],
+    [
+      "1fns",
+      "A"
+    ],
+    [
+      "1yy9",
+      "A"
+    ],
+    [
+      "4jqi",
+      "A"
+    ]
+  ]
+}

data/species.json

@@ -0,0 +1,3562 @@
+{
+  "4uuj": {
+    "chain": "C",
+    "organism": "STREPTOMYCES LIVIDANS",
+    "classification": "other"
+  },
+  "7amp": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8f5n": {
+    "chain": "A",
+    "organism": "Streptococcus pyogenes",
+    "classification": "bacterial"
+  },
+  "5bo1": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1ncc": {
+    "chain": "N",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "8x0t": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8g4t": {
+    "chain": "I",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "6o39": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3wfd": {
+    "chain": "C",
+    "organism": "Pseudomonas aeruginosa",
+    "classification": "bacterial"
+  },
+  "8w85": {
+    "chain": "H",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6rps": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "9ivk": {
+    "chain": "A",
+    "organism": "artificial sequences",
+    "classification": "other"
+  },
+  "7kql": {
+    "chain": "T",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7kyo": {
+    "chain": "B",
+    "organism": "Streptococcus pneumoniae serotype 2 (strain D39 / NCTC 7466)",
+    "classification": "bacterial"
+  },
+  "3o2d": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1uwx": {
+    "chain": "B",
+    "organism": "STREPTOCOCCUS SP.",
+    "classification": "bacterial"
+  },
+  "2oz4": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6xkq": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "8d9z": {
+    "chain": "D",
+    "organism": "Naja nivea",
+    "classification": "other"
+  },
+  "8v52": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4xnm": {
+    "chain": "C",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "6p67": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6qig": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "9dez": {
+    "chain": "C",
+    "organism": "Porcine deltacoronavirus",
+    "classification": "viral"
+  },
+  "8cdd": {
+    "chain": "B",
+    "organism": "Plasmodium falciparum 3D7",
+    "classification": "other"
+  },
+  "4qnp": {
+    "chain": "A",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "4lss": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4zs7": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "9b2w": {
+    "chain": "D",
+    "organism": "Human respirovirus 3",
+    "classification": "viral"
+  },
+  "6vgr": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4nnp": {
+    "chain": "B",
+    "organism": "Staphylococcus aureus subsp. aureus",
+    "classification": "bacterial"
+  },
+  "5lsp": {
+    "chain": "P",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7ec5": {
+    "chain": "B",
+    "organism": "Human enterovirus D68",
+    "classification": "viral"
+  },
+  "8ffe": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1mhp": {
+    "chain": "B",
+    "organism": "Rattus norvegicus",
+    "classification": "other"
+  },
+  "4uu9": {
+    "chain": "D",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "7tuf": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7sue": {
+    "chain": "D",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "7ul3": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8dk6": {
+    "chain": "E",
+    "organism": "Hepatitis C virus isolate HC-J6",
+    "classification": "viral"
+  },
+  "8y0q": {
+    "chain": "2",
+    "organism": "Foot-and-mouth disease virus O",
+    "classification": "viral"
+  },
+  "7upw": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus",
+    "classification": "viral"
+  },
+  "4qti": {
+    "chain": "U",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4o58": {
+    "chain": "A",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "4krp": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4yxl": {
+    "chain": "A",
+    "organism": "Mesocricetus auratus",
+    "classification": "other"
+  },
+  "7lr3": {
+    "chain": "C",
+    "organism": "Plasmodium berghei",
+    "classification": "other"
+  },
+  "4f2m": {
+    "chain": "E",
+    "organism": "TGEV virulent Purdue",
+    "classification": "other"
+  },
+  "5jxe": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8tzw": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6elu": {
+    "chain": "A",
+    "organism": "Trypanosoma brucei rhodesiense",
+    "classification": "other"
+  },
+  "6nyq": {
+    "chain": "C",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "9jbq": {
+    "chain": "C",
+    "organism": "Pseudomonas aeruginosa",
+    "classification": "bacterial"
+  },
+  "7wn8": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7quh": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7o9s": {
+    "chain": "A",
+    "organism": "Hantaan orthohantavirus",
+    "classification": "viral"
+  },
+  "8pg0": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1g9n": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "5bk2": {
+    "chain": "A",
+    "organism": "Escherichia coli",
+    "classification": "bacterial"
+  },
+  "7uvf": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6mei": {
+    "chain": "C",
+    "organism": "Hepacivirus C",
+    "classification": "viral"
+  },
+  "5w3l": {
+    "chain": "B",
+    "organism": "Human rhinovirus 14",
+    "classification": "viral"
+  },
+  "8tco": {
+    "chain": "C",
+    "organism": "Human betaherpesvirus 5",
+    "classification": "viral"
+  },
+  "3u9p": {
+    "chain": "C",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "6wio": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8vzn": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "5bv7": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7jkt": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7x28": {
+    "chain": "F",
+    "organism": "Middle East respiratory syndrome-related coronavirus",
+    "classification": "viral"
+  },
+  "6xxv": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3ld8": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4lst": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4d3c": {
+    "chain": "A",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "8a44": {
+    "chain": "A",
+    "organism": "Plasmodium vivax",
+    "classification": "other"
+  },
+  "4hwb": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3v6o": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6ogx": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2xra": {
+    "chain": "A",
+    "organism": "SYNTHETIC CONSTRUCT",
+    "classification": "other"
+  },
+  "6q20": {
+    "chain": "A",
+    "organism": "Influenza A virus (strain A/Japan/305/1957 H2N2)",
+    "classification": "viral"
+  },
+  "4jqi": {
+    "chain": "A",
+    "organism": "Rattus norvegicus",
+    "classification": "other"
+  },
+  "5ush": {
+    "chain": "X",
+    "organism": "Vaccinia virus",
+    "classification": "viral"
+  },
+  "3hae": {
+    "chain": "K",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2hmi": {
+    "chain": "B",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "3vg9": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6h2y": {
+    "chain": "D",
+    "organism": "Neisseria meningitidis",
+    "classification": "other"
+  },
+  "8u03": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6bkc": {
+    "chain": "E",
+    "organism": "Recombinant Hepatitis C virus HK6a/JFH-1",
+    "classification": "viral"
+  },
+  "6z3q": {
+    "chain": "C",
+    "organism": "Enterovirus A71",
+    "classification": "viral"
+  },
+  "8dn6": {
+    "chain": "A",
+    "organism": "Arabidopsis thaliana",
+    "classification": "other"
+  },
+  "4qci": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3d85": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3hi6": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8c7h": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7mrz": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7zqt": {
+    "chain": "E",
+    "organism": "Helicobacter pylori",
+    "classification": "other"
+  },
+  "7ued": {
+    "chain": "M",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6wgl": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1nsn": {
+    "chain": "S",
+    "organism": "Staphylococcus aureus",
+    "classification": "bacterial"
+  },
+  "8y31": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6pe8": {
+    "chain": "U",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1xiw": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7wg3": {
+    "chain": "K",
+    "organism": "Bos taurus",
+    "classification": "other"
+  },
+  "8dn7": {
+    "chain": "C",
+    "organism": "Pisum sativum",
+    "classification": "other"
+  },
+  "2hfg": {
+    "chain": "R",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5xez": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8v91": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2j4w": {
+    "chain": "D",
+    "organism": "PLASMODIUM VIVAX",
+    "classification": "other"
+  },
+  "8e8x": {
+    "chain": "1",
+    "organism": "Human poliovirus 3 strain Sabin",
+    "classification": "viral"
+  },
+  "1kb5": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "6gv4": {
+    "chain": "B",
+    "organism": "Human parechovirus 3",
+    "classification": "viral"
+  },
+  "8sxp": {
+    "chain": "C",
+    "organism": "Ophiophagus hannah",
+    "classification": "other"
+  },
+  "6ztr": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6cw3": {
+    "chain": "F",
+    "organism": "Saccharomyces cerevisiae S288c",
+    "classification": "fungal"
+  },
+  "8tq9": {
+    "chain": "B",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "6wfw": {
+    "chain": "G",
+    "organism": "Streptococcus sp. group G",
+    "classification": "bacterial"
+  },
+  "8ee0": {
+    "chain": "A",
+    "organism": "Saccharopolyspora erythraea",
+    "classification": "other"
+  },
+  "3lev": {
+    "chain": "A",
+    "organism": "Thermus aquaticus",
+    "classification": "other"
+  },
+  "6y9b": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7v23": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "8bk2": {
+    "chain": "B",
+    "organism": "Neisseria meningitidis serogroup B",
+    "classification": "other"
+  },
+  "5h35": {
+    "chain": "C",
+    "organism": "Sulfolobus solfataricus",
+    "classification": "other"
+  },
+  "5ldn": {
+    "chain": "A",
+    "organism": "Human adenovirus C serotype 5",
+    "classification": "viral"
+  },
+  "1ors": {
+    "chain": "C",
+    "organism": "Aeropyrum pernix",
+    "classification": "other"
+  },
+  "6fgb": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7tyv": {
+    "chain": "B",
+    "organism": "Lassa virus",
+    "classification": "viral"
+  },
+  "3qwo": {
+    "chain": "C",
+    "organism": "Staphylococcus aureus",
+    "classification": "bacterial"
+  },
+  "7n8q": {
+    "chain": "A",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4ye4": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus",
+    "classification": "viral"
+  },
+  "3uc0": {
+    "chain": "B",
+    "organism": "Dengue virus 4",
+    "classification": "viral"
+  },
+  "6mlk": {
+    "chain": "A",
+    "organism": "Saccharopolyspora erythraea",
+    "classification": "other"
+  },
+  "6hga": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1fns": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7n0u": {
+    "chain": "C",
+    "organism": "Betula pendula",
+    "classification": "other"
+  },
+  "8db4": {
+    "chain": "E",
+    "organism": "Arachis hypogaea",
+    "classification": "other"
+  },
+  "6id4": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6svl": {
+    "chain": "N",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4wfg": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7shu": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8k3c": {
+    "chain": "B",
+    "organism": "Henipavirus nipahense",
+    "classification": "viral"
+  },
+  "5e94": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8byu": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4cad": {
+    "chain": "I",
+    "organism": "METHANOCOCCUS MARIPALUDIS",
+    "classification": "other"
+  },
+  "4o9h": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5utz": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5eu7": {
+    "chain": "B",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "8e8r": {
+    "chain": "2",
+    "organism": "Human poliovirus 3 strain Sabin",
+    "classification": "viral"
+  },
+  "8uky": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "9dh2": {
+    "chain": "T",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7x8q": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8yxi": {
+    "chain": "A",
+    "organism": "Severe fever with thrombocytopenia syndrome virus",
+    "classification": "viral"
+  },
+  "5mo9": {
+    "chain": "X",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3skj": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3grw": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7xw6": {
+    "chain": "R",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6wzk": {
+    "chain": "E",
+    "organism": "Macaca fascicularis",
+    "classification": "other"
+  },
+  "8u2c": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6ks0": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7yms": {
+    "chain": "A",
+    "organism": "Coxsackievirus A16",
+    "classification": "viral"
+  },
+  "8j8r": {
+    "chain": "B",
+    "organism": "Bos taurus",
+    "classification": "other"
+  },
+  "6ba5": {
+    "chain": "N",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7daa": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1dee": {
+    "chain": "H",
+    "organism": "Staphylococcus aureus",
+    "classification": "bacterial"
+  },
+  "7ox3": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6was": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8tq8": {
+    "chain": "C",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "4yxk": {
+    "chain": "A",
+    "organism": "Cervus elaphus nelsoni",
+    "classification": "other"
+  },
+  "6u36": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2qqn": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3cvh": {
+    "chain": "M",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "4jpk": {
+    "chain": "A",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "1v7m": {
+    "chain": "V",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2ybr": {
+    "chain": "I",
+    "organism": "CENTRUROIDES NOXIUS HOFFMANN",
+    "classification": "other"
+  },
+  "4ht1": {
+    "chain": "T",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4bz2": {
+    "chain": "A",
+    "organism": "DENGUE VIRUS 4",
+    "classification": "viral"
+  },
+  "8hhy": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "5epm": {
+    "chain": "C",
+    "organism": "Ceratogyrus marshalli",
+    "classification": "other"
+  },
+  "7lxx": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "2zch": {
+    "chain": "P",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6nms": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6vep": {
+    "chain": "Q",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8tv3": {
+    "chain": "A",
+    "organism": "Borreliella burgdorferi B31",
+    "classification": "other"
+  },
+  "6m3b": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7zoz": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4rwy": {
+    "chain": "A",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "3u4e": {
+    "chain": "J",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4liq": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7rew": {
+    "chain": "I",
+    "organism": "Macaca fascicularis",
+    "classification": "other"
+  },
+  "8tlm": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8vsj": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6ks1": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4leo": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7s13": {
+    "chain": "C",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "6pi7": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1kyo": {
+    "chain": "P",
+    "organism": "Saccharomyces cerevisiae",
+    "classification": "fungal"
+  },
+  "8fdo": {
+    "chain": "C",
+    "organism": "Escherichia coli",
+    "classification": "bacterial"
+  },
+  "6u6u": {
+    "chain": "R",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8wre": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7s7i": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7lf8": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6phb": {
+    "chain": "I",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "8fgx": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3gbm": {
+    "chain": "A",
+    "organism": "Influenza A virus (A/Viet Nam/1203/2004(H5N1))",
+    "classification": "viral"
+  },
+  "2j5l": {
+    "chain": "A",
+    "organism": "PLASMODIUM FALCIPARUM",
+    "classification": "other"
+  },
+  "7ce2": {
+    "chain": "A",
+    "organism": "Clostridium tetani",
+    "classification": "other"
+  },
+  "6bf4": {
+    "chain": "A",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "2xqy": {
+    "chain": "E",
+    "organism": "SUID HERPESVIRUS",
+    "classification": "viral"
+  },
+  "7q0i": {
+    "chain": "D",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "6mug": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4yue": {
+    "chain": "C",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "4qww": {
+    "chain": "B",
+    "organism": "Bungarus fasciatus",
+    "classification": "other"
+  },
+  "6bfq": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8too": {
+    "chain": "I",
+    "organism": "Epstein-Barr virus",
+    "classification": "viral"
+  },
+  "7dm2": {
+    "chain": "A",
+    "organism": "Mycobacterium tuberculosis H37Rv",
+    "classification": "bacterial"
+  },
+  "8iuy": {
+    "chain": "A",
+    "organism": "H7N9 subtype",
+    "classification": "other"
+  },
+  "6wtu": {
+    "chain": "A",
+    "organism": "Plasmodium vivax (strain Salvador I)",
+    "classification": "other"
+  },
+  "8u1c": {
+    "chain": "A",
+    "organism": "Influenza B virus (B/Iowa/06/2017)",
+    "classification": "viral"
+  },
+  "7a0x": {
+    "chain": "B",
+    "organism": "Salmonella typhimurium (strain LT2 / SGSC1412 / ATCC 700720)",
+    "classification": "bacterial"
+  },
+  "7txt": {
+    "chain": "S",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3liz": {
+    "chain": "A",
+    "organism": "Blattella germanica",
+    "classification": "other"
+  },
+  "6ppg": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6nnf": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4ma7": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "8djg": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7kd6": {
+    "chain": "W",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4m1g": {
+    "chain": "B",
+    "organism": "Vaccinia virus",
+    "classification": "viral"
+  },
+  "8sic": {
+    "chain": "G",
+    "organism": "Human herpesvirus 4",
+    "classification": "viral"
+  },
+  "5tud": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6apb": {
+    "chain": "C",
+    "organism": "Human respiratory syncytial virus",
+    "classification": "viral"
+  },
+  "4a6y": {
+    "chain": "A",
+    "organism": "MUS MUSCULUS",
+    "classification": "other"
+  },
+  "1oaz": {
+    "chain": "A",
+    "organism": "ESCHERICHIA COLI",
+    "classification": "bacterial"
+  },
+  "5vic": {
+    "chain": "E",
+    "organism": "Dengue virus type 1 (strain Nauru/West Pac/1974)",
+    "classification": "viral"
+  },
+  "6k65": {
+    "chain": "A",
+    "organism": "Staphylococcus aureus (strain NCTC 8325)",
+    "classification": "bacterial"
+  },
+  "8r1d": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1lk3": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6meh": {
+    "chain": "C",
+    "organism": "Hepacivirus C",
+    "classification": "viral"
+  },
+  "7e7x": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "6wzl": {
+    "chain": "E",
+    "organism": "Macaca fascicularis",
+    "classification": "other"
+  },
+  "6j14": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4hc1": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1yy9": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8v2e": {
+    "chain": "C",
+    "organism": "synthetic construct",
+    "classification": "other"
+  },
+  "6wh9": {
+    "chain": "D",
+    "organism": "Saccharopolyspora erythraea",
+    "classification": "other"
+  },
+  "5hbv": {
+    "chain": "B",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "5wt9": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5fcu": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "6dkj": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6a3w": {
+    "chain": "L",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4xp9": {
+    "chain": "C",
+    "organism": "Drosophila melanogaster",
+    "classification": "other"
+  },
+  "4ogy": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7k7h": {
+    "chain": "A",
+    "organism": "Salmonella enterica subsp. enterica serovar Typhi str. CT18",
+    "classification": "bacterial"
+  },
+  "7rp2": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3q3g": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6mto": {
+    "chain": "T",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "5w4l": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7bsc": {
+    "chain": "A",
+    "organism": "Dengue virus 2",
+    "classification": "viral"
+  },
+  "5vpl": {
+    "chain": "A",
+    "organism": "Dermatophagoides farinae",
+    "classification": "other"
+  },
+  "4zso": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5e8d": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6umg": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2yss": {
+    "chain": "C",
+    "organism": "Gallus gallus",
+    "classification": "other"
+  },
+  "6xsw": {
+    "chain": "X",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6kz0": {
+    "chain": "J",
+    "organism": "Human rhinovirus 14",
+    "classification": "viral"
+  },
+  "6wzj": {
+    "chain": "E",
+    "organism": "Macaca fascicularis",
+    "classification": "other"
+  },
+  "4fqj": {
+    "chain": "A",
+    "organism": "Influenza B virus",
+    "classification": "viral"
+  },
+  "6uym": {
+    "chain": "E",
+    "organism": "Hepatitis C virus (isolate H)",
+    "classification": "viral"
+  },
+  "2xqb": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8oxv": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5b3j": {
+    "chain": "C",
+    "organism": "Rattus norvegicus",
+    "classification": "other"
+  },
+  "7ox1": {
+    "chain": "X",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "9dwe": {
+    "chain": "A",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "3lhp": {
+    "chain": "S",
+    "organism": "Artificial gene",
+    "classification": "other"
+  },
+  "8cz5": {
+    "chain": "A",
+    "organism": "Lagovirus",
+    "classification": "viral"
+  },
+  "7mjk": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "8txp": {
+    "chain": "A",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "6al5": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8sgi": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3l95": {
+    "chain": "X",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3tje": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3hmx": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7l0l": {
+    "chain": "E",
+    "organism": "Influenza A virus (A/Canada/720/2005(H2N2))",
+    "classification": "viral"
+  },
+  "8aci": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1nfd": {
+    "chain": "D",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "3vi4": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3wih": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7so5": {
+    "chain": "A",
+    "organism": "Clostridioides difficile R20291",
+    "classification": "other"
+  },
+  "3r1g": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1bgx": {
+    "chain": "T",
+    "organism": "Thermus aquaticus",
+    "classification": "other"
+  },
+  "9c44": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "6ewb": {
+    "chain": "C",
+    "organism": "Norovirus Hu/GII.4/Sydney/NSW0514/2012/AU",
+    "classification": "viral"
+  },
+  "5nh3": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4yqx": {
+    "chain": "M",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "9dx6": {
+    "chain": "A",
+    "organism": "Plasmodium vivax",
+    "classification": "other"
+  },
+  "5kn5": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3ru8": {
+    "chain": "X",
+    "organism": "Human immunodeficiency virus",
+    "classification": "viral"
+  },
+  "7lxw": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "3gbn": {
+    "chain": "A",
+    "organism": "Influenza A virus (A/Brevig Mission/1/1918(H1N1))",
+    "classification": "viral"
+  },
+  "7rt9": {
+    "chain": "Y",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6yio": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4ywg": {
+    "chain": "Q",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "3s35": {
+    "chain": "X",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6uyd": {
+    "chain": "F",
+    "organism": "Hepatitis C virus (isolate H)",
+    "classification": "viral"
+  },
+  "5u8r": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3kr3": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6a77": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7zxk": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6ion": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6wit": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7joo": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4tsa": {
+    "chain": "A",
+    "organism": "Gallus gallus",
+    "classification": "other"
+  },
+  "1ob1": {
+    "chain": "F",
+    "organism": "PLASMODIUM FALCIPARUM",
+    "classification": "other"
+  },
+  "6wzm": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1fj1": {
+    "chain": "E",
+    "organism": "Borrelia burgdorferi",
+    "classification": "other"
+  },
+  "7ttx": {
+    "chain": "A",
+    "organism": "Bat coronavirus RaTG13",
+    "classification": "viral"
+  },
+  "6wmw": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4ot1": {
+    "chain": "A",
+    "organism": "Human Cytomegalovirus",
+    "classification": "viral"
+  },
+  "7qu2": {
+    "chain": "C",
+    "organism": "Argentinian mammarenavirus",
+    "classification": "viral"
+  },
+  "7sgm": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7phu": {
+    "chain": "A",
+    "organism": "Plasmodium falciparum (isolate 3D7)",
+    "classification": "other"
+  },
+  "6sni": {
+    "chain": "X",
+    "organism": "Saccharomyces cerevisiae",
+    "classification": "fungal"
+  },
+  "6aod": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8e8y": {
+    "chain": "1",
+    "organism": "Human poliovirus 2 strain Sabin",
+    "classification": "viral"
+  },
+  "1e6j": {
+    "chain": "P",
+    "organism": "HIV-1 M\\:B_HXB2R",
+    "classification": "other"
+  },
+  "8gat": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5u3m": {
+    "chain": "A",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7ki6": {
+    "chain": "A",
+    "organism": "Hendra henipavirus",
+    "classification": "viral"
+  },
+  "2uzi": {
+    "chain": "R",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "4wv1": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8caf": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7ly0": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "5dhv": {
+    "chain": "M",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "5yoy": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4g6f": {
+    "chain": "F",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7xco": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "8vyn": {
+    "chain": "C",
+    "organism": "Human betaherpesvirus 5",
+    "classification": "viral"
+  },
+  "8jkf": {
+    "chain": "A",
+    "organism": "Zika virus",
+    "classification": "viral"
+  },
+  "6iek": {
+    "chain": "D",
+    "organism": "Rift valley fever virus",
+    "classification": "viral"
+  },
+  "6wbv": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6hx4": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7rxl": {
+    "chain": "F",
+    "organism": "Plasmodium falciparum (isolate 3D7)",
+    "classification": "other"
+  },
+  "5fb8": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5l0q": {
+    "chain": "A",
+    "organism": "Bos taurus",
+    "classification": "other"
+  },
+  "7a3o": {
+    "chain": "A",
+    "organism": "Dengue virus 1",
+    "classification": "viral"
+  },
+  "6vn1": {
+    "chain": "C",
+    "organism": "Human alphaherpesvirus 3",
+    "classification": "viral"
+  },
+  "8ol9": {
+    "chain": "H",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3mj9": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "6lz9": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5d93": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "3i50": {
+    "chain": "E",
+    "organism": "West Nile virus",
+    "classification": "viral"
+  },
+  "6blh": {
+    "chain": "G",
+    "organism": "Human respiratory syncytial virus A (strain rsb6256)",
+    "classification": "viral"
+  },
+  "8jel": {
+    "chain": "J",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5te4": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7m3n": {
+    "chain": "A",
+    "organism": "Canine parvovirus type 2",
+    "classification": "viral"
+  },
+  "6o3b": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5lcv": {
+    "chain": "A",
+    "organism": "Zika virus",
+    "classification": "viral"
+  },
+  "8ath": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5nmv": {
+    "chain": "K",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8hrx": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7uvi": {
+    "chain": "C",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "9fjk": {
+    "chain": "B",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "6lxj": {
+    "chain": "D",
+    "organism": "Influenza A virus (A/Anhui/1-BALF_RG44/2013(H7N9))",
+    "classification": "viral"
+  },
+  "7ttm": {
+    "chain": "A",
+    "organism": "Bat SARS-like coronavirus RsSHC014",
+    "classification": "viral"
+  },
+  "2aep": {
+    "chain": "A",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "1sy6": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8udz": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1xf5": {
+    "chain": "L",
+    "organism": "Finegoldia magna",
+    "classification": "other"
+  },
+  "1egj": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6xcj": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "6ml8": {
+    "chain": "A",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "5jq6": {
+    "chain": "A",
+    "organism": "Staphylococcus aureus",
+    "classification": "bacterial"
+  },
+  "4i2x": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6wo4": {
+    "chain": "E",
+    "organism": "Recombinant Hepatitis C virus HK6a/JFH-1",
+    "classification": "viral"
+  },
+  "4np4": {
+    "chain": "A",
+    "organism": "Clostridium difficile",
+    "classification": "other"
+  },
+  "4ps4": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3idx": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "6iuv": {
+    "chain": "A",
+    "organism": "Influenza A virus (A/Hong Kong/482/97(H5N1))",
+    "classification": "viral"
+  },
+  "8h07": {
+    "chain": "B",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "3gi9": {
+    "chain": "C",
+    "organism": "Methanocaldococcus jannaschii",
+    "classification": "other"
+  },
+  "6bgt": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8t9z": {
+    "chain": "A",
+    "organism": "Human metapneumovirus",
+    "classification": "viral"
+  },
+  "2vxs": {
+    "chain": "D",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "2vxt": {
+    "chain": "I",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "5u3d": {
+    "chain": "E",
+    "organism": "Finegoldia magna",
+    "classification": "other"
+  },
+  "6phc": {
+    "chain": "I",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "1fbi": {
+    "chain": "Y",
+    "organism": "Numida meleagris",
+    "classification": "other"
+  },
+  "6adb": {
+    "chain": "A",
+    "organism": "Escherichia coli (strain K12)",
+    "classification": "bacterial"
+  },
+  "4plj": {
+    "chain": "A",
+    "organism": "Hepatitis E virus",
+    "classification": "viral"
+  },
+  "1rjl": {
+    "chain": "C",
+    "organism": "Borrelia burgdorferi",
+    "classification": "other"
+  },
+  "5vyf": {
+    "chain": "F",
+    "organism": "Felis catus",
+    "classification": "other"
+  },
+  "4jzj": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5xj4": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8ruu": {
+    "chain": "Y",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6v4p": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4rdq": {
+    "chain": "C",
+    "organism": "Gallus gallus",
+    "classification": "other"
+  },
+  "4jlr": {
+    "chain": "C",
+    "organism": "Unknown",
+    "classification": "other"
+  },
+  "8tqa": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "8dcn": {
+    "chain": "F",
+    "organism": "Clostridioides difficile",
+    "classification": "other"
+  },
+  "6v4n": {
+    "chain": "W",
+    "organism": "Influenza B virus",
+    "classification": "viral"
+  },
+  "6vzi": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "5if0": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5k9k": {
+    "chain": "F",
+    "organism": "Influenza A virus (strain A/Hong Kong/1/1968 H3N2)",
+    "classification": "viral"
+  },
+  "5veb": {
+    "chain": "Y",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1tqb": {
+    "chain": "A",
+    "organism": "Ovis aries",
+    "classification": "other"
+  },
+  "3nh7": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4lvo": {
+    "chain": "A",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "8r8d": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4rgo": {
+    "chain": "S",
+    "organism": "Staphylococcus aureus",
+    "classification": "bacterial"
+  },
+  "7r58": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4ala": {
+    "chain": "C",
+    "organism": "DENGUE VIRUS 3",
+    "classification": "viral"
+  },
+  "4f37": {
+    "chain": "A",
+    "organism": "Escherichia coli",
+    "classification": "bacterial"
+  },
+  "7yru": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4aei": {
+    "chain": "C",
+    "organism": "ANDROCTONUS AUSTRALIS HECTOR",
+    "classification": "other"
+  },
+  "6u2f": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5w5x": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6q0l": {
+    "chain": "A",
+    "organism": "Influenza A virus (A/Beijing/262/1995(H1N1))",
+    "classification": "viral"
+  },
+  "5tq0": {
+    "chain": "A",
+    "organism": "Xenopus laevis",
+    "classification": "other"
+  },
+  "3jwd": {
+    "chain": "A",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "6oor": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "8g8d": {
+    "chain": "P",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7uot": {
+    "chain": "C",
+    "organism": "Lassa virus",
+    "classification": "viral"
+  },
+  "7lsg": {
+    "chain": "C",
+    "organism": "Tick-borne encephalitis virus",
+    "classification": "viral"
+  },
+  "8sak": {
+    "chain": "B",
+    "organism": "unclassified Merbecovirus",
+    "classification": "viral"
+  },
+  "6uvo": {
+    "chain": "D",
+    "organism": "Human respiratory syncytial virus A (strain A2)",
+    "classification": "viral"
+  },
+  "7uij": {
+    "chain": "D",
+    "organism": "Borreliella burgdorferi B31",
+    "classification": "other"
+  },
+  "7lse": {
+    "chain": "E",
+    "organism": "Tick-borne encephalitis virus Far Eastern subtype",
+    "classification": "viral"
+  },
+  "6cw2": {
+    "chain": "C",
+    "organism": "Saccharomyces cerevisiae",
+    "classification": "fungal"
+  },
+  "4j6r": {
+    "chain": "G",
+    "organism": "HUMAN IMMUNODEFICIENCY VIRUS 1",
+    "classification": "viral"
+  },
+  "7n3c": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "3wkm": {
+    "chain": "B",
+    "organism": "Aquifex aeolicus",
+    "classification": "other"
+  },
+  "6s5a": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5ikc": {
+    "chain": "N",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6s3d": {
+    "chain": "M",
+    "organism": "synthetic construct",
+    "classification": "other"
+  },
+  "7ahu": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8c7m": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7xnf": {
+    "chain": "A",
+    "organism": "Pangolin coronavirus",
+    "classification": "viral"
+  },
+  "8y0r": {
+    "chain": "2",
+    "organism": "Foot-and-mouth disease virus A",
+    "classification": "viral"
+  },
+  "8y6i": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3t2n": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4zff": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4m62": {
+    "chain": "T",
+    "organism": "synthetic construct",
+    "classification": "other"
+  },
+  "4ag4": {
+    "chain": "A",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "5ukr": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "8d1t": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4nzr": {
+    "chain": "M",
+    "organism": "Mycoplasma genitalium",
+    "classification": "other"
+  },
+  "6xlq": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8da1": {
+    "chain": "I",
+    "organism": "Bungarus multicinctus",
+    "classification": "other"
+  },
+  "7u5b": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3q1s": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6by3": {
+    "chain": "C",
+    "organism": "Streptomyces coelicolor",
+    "classification": "other"
+  },
+  "6lyn": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5occ": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8jxs": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4lmq": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5w5z": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7k9j": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "5k59": {
+    "chain": "B",
+    "organism": "Staphylococcus aureus (strain USA300)",
+    "classification": "bacterial"
+  },
+  "8kdm": {
+    "chain": "B",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "7lr4": {
+    "chain": "C",
+    "organism": "Plasmodium berghei",
+    "classification": "other"
+  },
+  "7lf7": {
+    "chain": "M",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8w86": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7wvg": {
+    "chain": "B",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "6wix": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7xrz": {
+    "chain": "Y",
+    "organism": "Escherichia coli",
+    "classification": "bacterial"
+  },
+  "1jrh": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3ks0": {
+    "chain": "A",
+    "organism": "Saccharomyces cerevisiae",
+    "classification": "fungal"
+  },
+  "7x29": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "5x2n": {
+    "chain": "C",
+    "organism": "Oryzias latipes",
+    "classification": "other"
+  },
+  "7lkf": {
+    "chain": "A",
+    "organism": "Gallus gallus",
+    "classification": "other"
+  },
+  "6mi2": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5dur": {
+    "chain": "A",
+    "organism": "Influenza A virus (A/Anhui/1/2005(H5N1))",
+    "classification": "viral"
+  },
+  "7kyl": {
+    "chain": "Z",
+    "organism": "Powassan virus",
+    "classification": "viral"
+  },
+  "7tuy": {
+    "chain": "R",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4qhu": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7e72": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4g7v": {
+    "chain": "S",
+    "organism": "Ciona intestinalis",
+    "classification": "other"
+  },
+  "8f38": {
+    "chain": "C",
+    "organism": "synthetic construct",
+    "classification": "other"
+  },
+  "8hpk": {
+    "chain": "A",
+    "organism": "Oxalobacter formigenes",
+    "classification": "other"
+  },
+  "6rlo": {
+    "chain": "J",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4dw2": {
+    "chain": "U",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6uj9": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8g3q": {
+    "chain": "G",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "4kxz": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8v5q": {
+    "chain": "G",
+    "organism": "Human alphaherpesvirus 3",
+    "classification": "viral"
+  },
+  "6z7z": {
+    "chain": "F",
+    "organism": "Sus scrofa",
+    "classification": "other"
+  },
+  "8vgn": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1ztx": {
+    "chain": "E",
+    "organism": "West Nile virus",
+    "classification": "viral"
+  },
+  "6jbt": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2r56": {
+    "chain": "A",
+    "organism": "Bos taurus",
+    "classification": "other"
+  },
+  "1iqd": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2q8a": {
+    "chain": "A",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "6pis": {
+    "chain": "B",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "5cbe": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4xwo": {
+    "chain": "A",
+    "organism": "Saccharomyces cerevisiae (ATCC 204508 / S288c)",
+    "classification": "fungal"
+  },
+  "7upb": {
+    "chain": "D",
+    "organism": "Nipah henipavirus",
+    "classification": "viral"
+  },
+  "6uyf": {
+    "chain": "E",
+    "organism": "Recombinant Hepatitis C virus HK6a/JFH-1",
+    "classification": "viral"
+  },
+  "8tbq": {
+    "chain": "P",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7ox4": {
+    "chain": "C",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "6j5d": {
+    "chain": "A",
+    "organism": "Louping ill virus",
+    "classification": "viral"
+  },
+  "4dtg": {
+    "chain": "K",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6cmi": {
+    "chain": "B",
+    "organism": "Hendra virus",
+    "classification": "viral"
+  },
+  "3efd": {
+    "chain": "K",
+    "organism": "Escherichia coli",
+    "classification": "bacterial"
+  },
+  "4k3j": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8rp8": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7a3q": {
+    "chain": "A",
+    "organism": "Dengue virus 4",
+    "classification": "viral"
+  },
+  "1uac": {
+    "chain": "Y",
+    "organism": "Meleagris gallopavo",
+    "classification": "other"
+  },
+  "7usl": {
+    "chain": "C",
+    "organism": "Bordetella pertussis",
+    "classification": "other"
+  },
+  "6kyz": {
+    "chain": "A",
+    "organism": "Human rhinovirus 14",
+    "classification": "viral"
+  },
+  "3ehb": {
+    "chain": "B",
+    "organism": "Paracoccus denitrificans",
+    "classification": "other"
+  },
+  "4u6v": {
+    "chain": "B",
+    "organism": "Staphylococcus aureus subsp. aureus TCH60",
+    "classification": "bacterial"
+  },
+  "5i9q": {
+    "chain": "A",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "1wej": {
+    "chain": "F",
+    "organism": "Equus caballus",
+    "classification": "other"
+  },
+  "8xi6": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "5te7": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7nx3": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7q6c": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4xmp": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4dkf": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7jtg": {
+    "chain": "E",
+    "organism": "Recombinant Hepatitis C virus HK6a/JFH-1",
+    "classification": "viral"
+  },
+  "5d8j": {
+    "chain": "A",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "7xy8": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5ob5": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6iap": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6vvu": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8vvm": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6ktr": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6iw2": {
+    "chain": "D",
+    "organism": "Yellow fever virus (strain 17D vaccine)",
+    "classification": "viral"
+  },
+  "6otc": {
+    "chain": "A",
+    "organism": "Lake Victoria marburgvirus (strain Popp-67)",
+    "classification": "viral"
+  },
+  "5otj": {
+    "chain": "D",
+    "organism": "Phleum pratense",
+    "classification": "other"
+  },
+  "6mej": {
+    "chain": "C",
+    "organism": "Hepacivirus C",
+    "classification": "viral"
+  },
+  "3l5w": {
+    "chain": "J",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4okv": {
+    "chain": "F",
+    "organism": "Anopheles stephensi",
+    "classification": "other"
+  },
+  "6p9h": {
+    "chain": "A",
+    "organism": "Staphylococcus aureus",
+    "classification": "bacterial"
+  },
+  "6hig": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8zd5": {
+    "chain": "D",
+    "organism": "Canis lupus familiaris",
+    "classification": "other"
+  },
+  "8ulj": {
+    "chain": "B",
+    "organism": "Respiratory syncytial virus A2",
+    "classification": "viral"
+  },
+  "5th9": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3pnw": {
+    "chain": "R",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5vjo": {
+    "chain": "F",
+    "organism": "Anas platyrhynchos",
+    "classification": "other"
+  },
+  "8j80": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5l6y": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6cxy": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1jhl": {
+    "chain": "A",
+    "organism": "Phasianus colchicus",
+    "classification": "other"
+  },
+  "1nl0": {
+    "chain": "G",
+    "organism": "Unknown",
+    "classification": "other"
+  },
+  "3l5x": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8vvk": {
+    "chain": "B",
+    "organism": "Crimean-Congo hemorrhagic fever virus",
+    "classification": "viral"
+  },
+  "7tpd": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6j15": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7zli": {
+    "chain": "A",
+    "organism": "Caenorhabditis elegans",
+    "classification": "other"
+  },
+  "2h9g": {
+    "chain": "R",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3lh2": {
+    "chain": "V",
+    "organism": "ARTIFICIAL GENE",
+    "classification": "other"
+  },
+  "8dao": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4i3r": {
+    "chain": "G",
+    "organism": "Human Immunodeficiency Virus",
+    "classification": "viral"
+  },
+  "7vgr": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "3thm": {
+    "chain": "F",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8vgm": {
+    "chain": "A",
+    "organism": "Aliarcobacter butzleri RM4018",
+    "classification": "other"
+  },
+  "8dfh": {
+    "chain": "A",
+    "organism": "Plasmodium falciparum 3D7",
+    "classification": "other"
+  },
+  "7wtf": {
+    "chain": "D",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "8y6h": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5mhr": {
+    "chain": "D",
+    "organism": "Reovirus sp.",
+    "classification": "viral"
+  },
+  "8h3n": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "7ujd": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8d9y": {
+    "chain": "K",
+    "organism": "Oxyuranus scutellatus scutellatus",
+    "classification": "other"
+  },
+  "1uj3": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3nfp": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6o9i": {
+    "chain": "C",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "4k94": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7bbj": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7mlh": {
+    "chain": "F",
+    "organism": "Dermatophagoides pteronyssinus",
+    "classification": "other"
+  },
+  "7o52": {
+    "chain": "U",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6umx": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7rah": {
+    "chain": "E",
+    "organism": "Bordetella pertussis",
+    "classification": "other"
+  },
+  "5e8e": {
+    "chain": "H",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4edw": {
+    "chain": "V",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7t0l": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8y6a": {
+    "chain": "B",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "4irz": {
+    "chain": "A",
+    "organism": "Oryctolagus cuniculus",
+    "classification": "other"
+  },
+  "4lu5": {
+    "chain": "A",
+    "organism": "Vaccinia virus",
+    "classification": "viral"
+  },
+  "6sv2": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8tui": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8dyx": {
+    "chain": "I",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "8w83": {
+    "chain": "L",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4rrp": {
+    "chain": "R",
+    "organism": "Saccharomyces cerevisiae",
+    "classification": "fungal"
+  },
+  "5gzn": {
+    "chain": "E",
+    "organism": "Zika virus",
+    "classification": "viral"
+  },
+  "5tl5": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4ypg": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5kw9": {
+    "chain": "A",
+    "organism": "Norwalk virus",
+    "classification": "viral"
+  },
+  "2q8b": {
+    "chain": "A",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "8a1e": {
+    "chain": "A",
+    "organism": "Rabies virus strain Pasteur vaccin",
+    "classification": "viral"
+  },
+  "4d9q": {
+    "chain": "A",
+    "organism": "Macaca mulatta",
+    "classification": "other"
+  },
+  "6wo5": {
+    "chain": "F",
+    "organism": "Hepatitis C virus (isolate H)",
+    "classification": "viral"
+  },
+  "2nyy": {
+    "chain": "A",
+    "organism": "Clostridium botulinum",
+    "classification": "other"
+  },
+  "7lfa": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6cyf": {
+    "chain": "Q",
+    "organism": "Pseudomonas aeruginosa",
+    "classification": "bacterial"
+  },
+  "4zfg": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8ts0": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6n6b": {
+    "chain": "A",
+    "organism": "Influenza A virus (A/Minnesota/11/2010(H3N2))",
+    "classification": "viral"
+  },
+  "6lxi": {
+    "chain": "B",
+    "organism": "Influenza A virus (strain A/Brevig Mission/1/1918 H1N1)",
+    "classification": "viral"
+  },
+  "7phw": {
+    "chain": "D",
+    "organism": "Plasmodium falciparum (isolate 3D7)",
+    "classification": "other"
+  },
+  "5vcn": {
+    "chain": "A",
+    "organism": "Dermatophagoides pteronyssinus",
+    "classification": "other"
+  },
+  "5n7w": {
+    "chain": "X",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4uta": {
+    "chain": "B",
+    "organism": "DENGUE VIRUS 2",
+    "classification": "viral"
+  },
+  "3zkn": {
+    "chain": "B",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "3w9e": {
+    "chain": "C",
+    "organism": "Human herpesvirus 2",
+    "classification": "viral"
+  },
+  "8vdl": {
+    "chain": "C",
+    "organism": "Plasmodium falciparum HB3",
+    "classification": "other"
+  },
+  "3u2s": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "7z2m": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5vkd": {
+    "chain": "A",
+    "organism": "Bundibugyo ebolavirus",
+    "classification": "viral"
+  },
+  "6xqw": {
+    "chain": "E",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "6ck9": {
+    "chain": "G",
+    "organism": "Human immunodeficiency virus 1",
+    "classification": "viral"
+  },
+  "4xvu": {
+    "chain": "H",
+    "organism": "Saccharomyces cerevisiae (ATCC 204508 / S288c)",
+    "classification": "fungal"
+  },
+  "4oii": {
+    "chain": "B",
+    "organism": "West Nile virus",
+    "classification": "viral"
+  },
+  "8gkl": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6i8s": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7unb": {
+    "chain": "R",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "8dg9": {
+    "chain": "C",
+    "organism": "Respiratory syncytial virus A2",
+    "classification": "viral"
+  },
+  "7zjl": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "7x8p": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5wk3": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6vtw": {
+    "chain": "A",
+    "organism": "Human respiratory syncytial virus A2",
+    "classification": "viral"
+  },
+  "8tp7": {
+    "chain": "C",
+    "organism": "Influenza A virus (A/Singapore/1/1957(H2N2))",
+    "classification": "viral"
+  },
+  "6u9s": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8slb": {
+    "chain": "A",
+    "organism": "Thermotoga maritima MSB8",
+    "classification": "other"
+  },
+  "2adf": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2vxq": {
+    "chain": "A",
+    "organism": "PHLEUM PRATENSE",
+    "classification": "other"
+  },
+  "4jr9": {
+    "chain": "A",
+    "organism": "Escherichia coli",
+    "classification": "bacterial"
+  },
+  "7kpj": {
+    "chain": "E",
+    "organism": "Ruminococcus gnavus",
+    "classification": "other"
+  },
+  "8jnk": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7qu1": {
+    "chain": "C",
+    "organism": "Machupo mammarenavirus",
+    "classification": "viral"
+  },
+  "8urf": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6u8c": {
+    "chain": "B",
+    "organism": "Streptococcus sp. 'group G'",
+    "classification": "bacterial"
+  },
+  "6tyb": {
+    "chain": "G",
+    "organism": "Simian immunodeficiency virus",
+    "classification": "viral"
+  },
+  "7q6e": {
+    "chain": "A",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "5o1r": {
+    "chain": "A",
+    "organism": "Neisseria meningitidis",
+    "classification": "other"
+  },
+  "7ket": {
+    "chain": "C",
+    "organism": "Neisseria meningitidis",
+    "classification": "other"
+  },
+  "8t03": {
+    "chain": "B",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "8da0": {
+    "chain": "E",
+    "organism": "Dendroaspis polylepis polylepis",
+    "classification": "other"
+  },
+  "7kq7": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2jel": {
+    "chain": "P",
+    "organism": "Escherichia coli",
+    "classification": "bacterial"
+  },
+  "8tfl": {
+    "chain": "B",
+    "organism": "Ricinus communis",
+    "classification": "other"
+  },
+  "5w2b": {
+    "chain": "A",
+    "organism": "Reston ebolavirus",
+    "classification": "viral"
+  },
+  "6s8j": {
+    "chain": "E",
+    "organism": "Ebola virus",
+    "classification": "viral"
+  },
+  "3b9k": {
+    "chain": "F",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "4khx": {
+    "chain": "A",
+    "organism": "Human Immunodeficiency Virus 1",
+    "classification": "viral"
+  },
+  "8tfn": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8d7e": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "9gwt": {
+    "chain": "P",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6n81": {
+    "chain": "A",
+    "organism": "Norovirus Hu/GII.4/Farmington Hills/2004/USA",
+    "classification": "viral"
+  },
+  "4mwf": {
+    "chain": "D",
+    "organism": "Hepatitis C virus (isolate H)",
+    "classification": "viral"
+  },
+  "7chz": {
+    "chain": "I",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3mxw": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2ypv": {
+    "chain": "A",
+    "organism": "NEISSERIA MENINGITIDIS MC58",
+    "classification": "other"
+  },
+  "8ahn": {
+    "chain": "A",
+    "organism": "Sin Nombre orthohantavirus",
+    "classification": "viral"
+  },
+  "5b71": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7c61": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7uvs": {
+    "chain": "C",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "8f60": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "1yjd": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6m58": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6osv": {
+    "chain": "K",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8jlw": {
+    "chain": "A",
+    "organism": "Crimean-Congo hemorrhagic fever orthonairovirus",
+    "classification": "viral"
+  },
+  "1cl7": {
+    "chain": "I",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "3rkd": {
+    "chain": "A",
+    "organism": "Hepatitis E virus",
+    "classification": "viral"
+  },
+  "5wi9": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8y3u": {
+    "chain": "K",
+    "organism": "Ebola virus",
+    "classification": "viral"
+  },
+  "6mvl": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7l7r": {
+    "chain": "G",
+    "organism": "Crimean-Congo hemorrhagic fever virus (strain Nigeria/IbAr10200/1970)",
+    "classification": "viral"
+  },
+  "8jlx": {
+    "chain": "A",
+    "organism": "Crimean-Congo hemorrhagic fever orthonairovirus",
+    "classification": "viral"
+  },
+  "8b7h": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2qqk": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5d72": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8tv1": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5ggv": {
+    "chain": "Y",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5y9j": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6c9u": {
+    "chain": "A",
+    "organism": "Saccharopolyspora erythraea",
+    "classification": "other"
+  },
+  "8ut3": {
+    "chain": "C",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "5wux": {
+    "chain": "E",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6uyn": {
+    "chain": "A",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  },
+  "4ij3": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8a99": {
+    "chain": "C",
+    "organism": "Severe acute respiratory syndrome coronavirus 2",
+    "classification": "viral"
+  },
+  "6ddm": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "4ffv": {
+    "chain": "B",
+    "organism": "Rattus norvegicus",
+    "classification": "other"
+  },
+  "2bdn": {
+    "chain": "A",
+    "organism": "Unknown",
+    "classification": "other"
+  },
+  "6h3t": {
+    "chain": "B",
+    "organism": "Bovine Schmallenberg virus",
+    "classification": "viral"
+  },
+  "6x3x": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7yk4": {
+    "chain": "B",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6qb6": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2wuc": {
+    "chain": "A",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "6gku": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5eii": {
+    "chain": "I",
+    "organism": "Saccharomyces cerevisiae",
+    "classification": "fungal"
+  },
+  "6k7o": {
+    "chain": "P",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8e8s": {
+    "chain": "2",
+    "organism": "Poliovirus 2",
+    "classification": "viral"
+  },
+  "2j88": {
+    "chain": "A",
+    "organism": "APIS MELLIFERA",
+    "classification": "other"
+  },
+  "4cni": {
+    "chain": "D",
+    "organism": "HOMO SAPIENS",
+    "classification": "human"
+  },
+  "7cj2": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "3u30": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8qya": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "2arj": {
+    "chain": "R",
+    "organism": "Mus musculus",
+    "classification": "other"
+  },
+  "7df1": {
+    "chain": "D",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7dc8": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7sem": {
+    "chain": "F",
+    "organism": "Human metapneumovirus",
+    "classification": "viral"
+  },
+  "5f3b": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8w84": {
+    "chain": "C",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8f5i": {
+    "chain": "A",
+    "organism": "Rhodopseudomonas palustris",
+    "classification": "bacterial"
+  },
+  "5xxy": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6e63": {
+    "chain": "A",
+    "organism": "Plasmodium falciparum",
+    "classification": "other"
+  },
+  "3nps": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "8yx9": {
+    "chain": "J",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "7vux": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "5hys": {
+    "chain": "G",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "9ima": {
+    "chain": "A",
+    "organism": "Homo sapiens",
+    "classification": "human"
+  },
+  "6dfj": {
+    "chain": "E",
+    "organism": "Dengue virus 1",
+    "classification": "viral"
+  },
+  "6flb": {
+    "chain": "G",
+    "organism": "Dengue virus 2",
+    "classification": "viral"
+  },
+  "4f15": {
+    "chain": "J",
+    "organism": "Influenza A virus",
+    "classification": "viral"
+  }
+}

models/ReCEP/20250626_110438/best_mcc_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1358ed202985dce98a889affa264ff3065bfccc942d7b0043e6f3b0700eb8ea0
+size 19059594

models/ReCEP/20250626_110438/config.json

@@ -0,0 +1,70 @@
+{
+  "model_config": {
+    "in_dim": 2560,
+    "rsa": true,
+    "dihedral": true,
+    "node_dims": [
+      512,
+      256,
+      256
+    ],
+    "edge_dim": 32,
+    "dropout": 0.4,
+    "activation": "gelu",
+    "residual": true,
+    "attention": true,
+    "normalize": true,
+    "coords_agg": "mean",
+    "ffn": true,
+    "batch_norm": true,
+    "concat": true,
+    "addition": false,
+    "pooling": "attention",
+    "fusion_type": "concat",
+    "node_gate": false,
+    "node_norm": true,
+    "node_layers": 2,
+    "out_dropout": 0.2,
+    "use_egnn": true,
+    "encoder": "esmc"
+  },
+  "training_config": {
+    "num_epoch": 120,
+    "batch_size": 64,
+    "lr": 5e-05,
+    "weight_decay": 1e-05,
+    "patience": 15,
+    "threshold": 0.5,
+    "mixed_precision": false,
+    "device_id": 0
+  },
+  "data_config": {
+    "radii": [
+      16,
+      18,
+      20
+    ],
+    "zero_ratio": 0.3,
+    "undersample": 0.5,
+    "seed": 42
+  },
+  "loss_config": {
+    "region_loss_type": "mse",
+    "reg_weight": 10.0,
+    "cls_type": "bce",
+    "gamma_high_cls": 2.0,
+    "regression_type": "smooth_l1",
+    "node_loss_type": "focal",
+    "alpha": 2.0,
+    "gamma": 2.0,
+    "pos_weight": 8.0,
+    "node_loss_weight": 0.5,
+    "region_weight": 1.0,
+    "consistency_weight": 0.3,
+    "consistency_type": "mse",
+    "label_smoothing": 0.1,
+    "gradnorm": true,
+    "gradnorm_alpha": 2.0,
+    "gradnorm_update_freq": 10
+  }
+}

requirements.txt

@@ -0,0 +1,49 @@
+# requirements.txt - For Hugging Face Spaces
+numpy==1.26.4
+scipy==1.13.0
+pandas==2.1.4
+scikit-learn==1.5.2
+
+torch==2.5.0
+torchvision==0.20.0
+torchaudio==2.5.0
+
+torch-geometric==2.6.1
+--find-links https://data.pyg.org/whl/torch-2.5.0+cpu.html
+torch-scatter
+torch-sparse
+torch-cluster
+torch-spline-conv
+
+torchtext==0.18.0
+torchmetrics==1.6.0
+torch-optimi==0.2.1
+torch-optimizer==0.3.0
+
+esm==3.1.3
+
+biopython==1.85
+biotite==0.41.2
+
+h5py==3.9.0
+py3Dmol==2.4.2
+matplotlib==3.9.3
+seaborn==0.13.2
+tqdm==4.67.1
+prettytable==3.16.0
+
+gradio==4.44.1
+fastapi==0.104.1
+uvicorn==0.24.0
+python-multipart>=0.0.9
+jinja2==3.1.2
+aiofiles==23.2.1
+
+cloudpathlib==0.16.0
+requests==2.31.0
+pillow==10.1.0
+
+networkx>=2.8
+
+notebook==7.3.2
+ipykernel==6.29.5

src/bce.egg-info/PKG-INFO

@@ -0,0 +1,3 @@
+Metadata-Version: 2.4
+Name: bce
+Version: 0.1

src/bce.egg-info/SOURCES.txt

@@ -0,0 +1,43 @@
+README.md
+setup.py
+src/bce/__init__.py
+src/bce/check_surf.py
+src/bce/extract_embeddings.py
+src/bce/loss.py
+src/bce/prediction_tools.py
+src/bce.egg-info/PKG-INFO
+src/bce.egg-info/SOURCES.txt
+src/bce.egg-info/dependency_links.txt
+src/bce.egg-info/top_level.txt
+src/bce/antigen/__init__.py
+src/bce/antigen/antigen.py
+src/bce/antigen/pc.py
+src/bce/data/__init__.py
+src/bce/data/data.py
+src/bce/data/data_.py
+src/bce/data/raw_data_generation.py
+src/bce/data/utils.py
+src/bce/data/utils_.py
+src/bce/model/EGNN.py
+src/bce/model/ReGEP.py
+src/bce/model/__init__.py
+src/bce/model/activation.py
+src/bce/model/baseline.py
+src/bce/model/dihedral.py
+src/bce/model/pooling.py
+src/bce/model/scheduler.py
+src/bce/utils/__init__.py
+src/bce/utils/check_struct.py
+src/bce/utils/constants.py
+src/bce/utils/data_tools.py
+src/bce/utils/dssp.py
+src/bce/utils/egnn_trainer.py
+src/bce/utils/evaluatror.py
+src/bce/utils/loading.py
+src/bce/utils/metrics.py
+src/bce/utils/results.py
+src/bce/utils/sequence.py
+src/bce/utils/str_data_tools.py
+src/bce/utils/tools.py
+src/bce/utils/trainer.py
+src/bce/utils/training_tools.py

src/bce.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

src/bce.egg-info/top_level.txt

@@ -0,0 +1 @@
+bce

src/bce/antigen/pc.py

@@ -0,0 +1,37 @@
+# Amino acid conversion mappings
+AMINO_ACID_1TO3 = {
+    'A': 'ALA', 'R': 'ARG', 'N': 'ASN', 'D': 'ASP', 'C': 'CYS',
+    'Q': 'GLN', 'E': 'GLU', 'G': 'GLY', 'H': 'HIS', 'I': 'ILE',
+    'L': 'LEU', 'K': 'LYS', 'M': 'MET', 'F': 'PHE', 'P': 'PRO',
+    'S': 'SER', 'T': 'THR', 'W': 'TRP', 'Y': 'TYR', 'V': 'VAL'
+}
+
+AMINO_ACID_3TO1 = {
+    'ALA': 'A', 'ARG': 'R', 'ASN': 'N', 'ASP': 'D', 'CYS': 'C',
+    'GLN': 'Q', 'GLU': 'E', 'GLY': 'G', 'HIS': 'H', 'ILE': 'I',
+    'LEU': 'L', 'LYS': 'K', 'MET': 'M', 'PHE': 'F', 'PRO': 'P',
+    'SER': 'S', 'THR': 'T', 'TRP': 'W', 'TYR': 'Y', 'VAL': 'V'
+}
+
+MAX_ASA = {
+    'ALA': 129.0,
+    'ARG': 274.0,
+    'ASN': 195.0,
+    'ASP': 193.0,
+    'CYS': 167.0,
+    'GLN': 223.0,
+    'GLU': 225.0,
+    'GLY': 104.0,
+    'HIS': 224.0,
+    'ILE': 197.0,
+    'LEU': 201.0,
+    'LYS': 236.0,
+    'MET': 224.0,
+    'PHE': 240.0,
+    'PRO': 159.0,
+    'SER': 155.0,
+    'THR': 172.0,
+    'TRP': 285.0,
+    'TYR': 263.0,
+    'VAL': 174.0,
+}

src/bce/data/data.py

@@ -0,0 +1,1262 @@
+import os
+import h5py
+import torch
+import numpy as np
+import json
+import random
+from pathlib import Path
+from multiprocessing import Pool
+
+from typing import List, Tuple, Dict, Optional, Union
+from torch_geometric.data import Data, Dataset, Batch
+from torch_geometric.loader import DataLoader
+from tqdm import tqdm
+import pickle
+
+from ..utils.loading import load_epitopes_csv, load_data_split
+from ..utils.constants import BASE_DIR
+from .utils import create_graph_data, create_graph_data_full
+
+
+def apply_undersample(data_list: List, undersample_param: Union[int, float], seed: int = 42, verbose: bool = True):
+    """
+    Apply undersampling to a data list.
+    
+    Args:
+        data_list: List of data samples
+        undersample_param: If int, sample that many samples; if float (0-1), sample that fraction of data
+        seed: Random seed for reproducibility
+        verbose: Whether to print sampling information
+        
+    Returns:
+        Undersampled data list
+    """
+    if undersample_param is None:
+        return data_list
+    
+    original_size = len(data_list)
+    
+    if isinstance(undersample_param, float):
+        # Sample a fraction of the data
+        if not (0 < undersample_param <= 1.0):
+            raise ValueError(f"Float undersample must be between 0 and 1, got {undersample_param}")
+        target_size = int(len(data_list) * undersample_param)
+    elif isinstance(undersample_param, int):
+        # Sample a specific number of samples
+        if undersample_param <= 0:
+            raise ValueError(f"Int undersample must be positive, got {undersample_param}")
+        target_size = min(undersample_param, len(data_list))
+    else:
+        raise ValueError(f"Undersample must be int, float, or None, got {type(undersample_param)}")
+    
+    if target_size < len(data_list):
+        # Set random seed for reproducibility
+        random.seed(seed)
+        sampled_data = random.sample(data_list, target_size)
+        
+        if verbose:
+            print(f"Applied undersampling: {original_size} -> {target_size} samples")
+        
+        return sampled_data
+    elif verbose:
+        print(f"No undersampling applied: requested {target_size}, available {original_size}")
+    
+    return data_list
+
+class AntigenDataset(Dataset):
+    """
+    Dataset for antigen chains.
+    Each data point represents a complete protein as a graph, with nodes being residues
+    and edges based on spatial distance (< 18 Å).
+    """
+    def __init__(
+        self,
+        data_split: str = "train",
+        radius: float = 18,
+        threshold: float = 0.25,
+        num_posenc: int = 16,
+        num_rbf: int = 16,
+        undersample: Union[int, float, None] = None,
+        cache_dir: Optional[str] = None,
+        force_rebuild: bool = False,
+        verbose: bool = True,
+        seed: int = 42,
+        encoder: str = "esmc"
+    ):
+        """
+        Initialize the antigen dataset.
+        
+        Args:
+            data_split: Data split name ('train', 'val', 'test')
+            radius: Distance threshold for edge creation (Å)
+            threshold: SASA threshold for surface residues (not used in full protein)
+            num_posenc: Number of positional encoding features
+            num_rbf: Number of RBF features
+            undersample: Undersample parameter (int for count, float for ratio)
+            cache_dir: Directory to cache processed data
+            force_rebuild: Whether to force rebuild the dataset
+            verbose: Whether to print progress information
+            seed: Random seed for reproducibility
+            encoder: Encoder type ('esmc' or 'esm2')
+        """
+        self.data_split = data_split
+        self.radius = radius
+        self.threshold = threshold
+        self.num_posenc = num_posenc
+        self.num_rbf = num_rbf
+        self.undersample = undersample
+        self.verbose = verbose
+        self.seed = seed
+        self.encoder = encoder
+        
+        # Set cache directory
+        if cache_dir is None:
+            cache_dir = Path(f"{BASE_DIR}/data/full_region_cache/antigen_r{radius}")
+        self.cache_dir = Path(cache_dir)
+        self.cache_dir.mkdir(parents=True, exist_ok=True)
+        
+        # Cache file for this configuration
+        self.cache_file = self.cache_dir / f"{data_split}_antigen_dataset.h5"
+        
+        # Load data splits and epitope information
+        self.antigens = load_data_split(data_split, verbose=verbose)
+        _, _, self.epitope_dict = load_epitopes_csv()
+        
+        # Initialize data list
+        self.data_list = []
+        
+        # Load or build dataset
+        if self.cache_file.exists() and not force_rebuild:
+            if verbose:
+                print(f"Loading cached antigen dataset from {self.cache_file}")
+            self._load_cache()
+        else:
+            if verbose:
+                print(f"Building antigen dataset for {data_split} split...")
+            self._build_dataset()
+            self._save_cache()
+        
+        super().__init__()
+    
+    def _load_protein_data(self, pdb_id: str, chain_id: str) -> Optional[Dict]:
+        """
+        Load precomputed protein data from files.
+        
+        Args:
+            pdb_id: PDB ID
+            chain_id: Chain ID
+            
+        Returns:
+            Dictionary containing all protein data or None if loading fails
+        """
+        try:
+            protein_key = f"{pdb_id}_{chain_id}"
+            
+            # Load embeddings
+            embedding_file = Path(BASE_DIR) / "data" / "embeddings" / self.encoder / f"{protein_key}.h5"
+            if not embedding_file.exists():
+                if self.verbose:
+                    print(f"Embedding file not found: {embedding_file}")
+                return None
+            
+            with h5py.File(embedding_file, "r") as h5f:
+                embeddings = h5f["embedding"][:]
+            
+            # Load backbone atoms
+            coords_file = Path(BASE_DIR) / "data" / "coords" / f"{protein_key}.npy"
+            if not coords_file.exists():
+                if self.verbose:
+                    print(f"Coords file not found: {coords_file}")
+                return None
+            backbone_atoms = np.load(coords_file)
+            
+            # Load RSA values
+            rsa_file = Path(BASE_DIR) / "data" / "rsa" / f"{protein_key}.npy"
+            if not rsa_file.exists():
+                if self.verbose:
+                    print(f"RSA file not found: {rsa_file}")
+                return None
+            rsa_values = np.load(rsa_file)
+            
+            # Load epitope data from epitope_dict
+            binary_labels = self.epitope_dict.get(protein_key, [])
+            
+            # Create epitope indices from binary labels
+            epitope_indices = []
+            for idx, is_epitope in enumerate(binary_labels):
+                if is_epitope == 1:
+                    epitope_indices.append(idx)
+            
+            return {
+                'embeddings': embeddings,
+                'backbone_atoms': backbone_atoms,
+                'rsa_values': rsa_values,
+                'epitope_indices': epitope_indices,
+            }
+            
+        except Exception as e:
+            if self.verbose:
+                print(f"Error loading protein data for {pdb_id}_{chain_id}: {str(e)}")
+            return None
+    
+    def _build_dataset(self):
+        """Build the dataset from precomputed data files."""
+        failed_proteins = []
+        
+        for pdb_id, chain_id in tqdm(self.antigens, desc=f"Processing {self.data_split} antigens", 
+                                   disable=not self.verbose):
+            try:
+                # Load precomputed data
+                protein_data = self._load_protein_data(pdb_id, chain_id)
+                if protein_data is None:
+                    failed_proteins.append(f"{pdb_id}_{chain_id}")
+                    continue
+                
+                embeddings = protein_data['embeddings']
+                backbone_atoms = protein_data['backbone_atoms']
+                rsa_values = protein_data['rsa_values']
+                epitope_indices = protein_data['epitope_indices']
+                
+                # Create graph data for the full protein
+                graph_data = create_graph_data_full(
+                    embeddings=embeddings,
+                    backbone_atoms=backbone_atoms,
+                    rsa_values=rsa_values,
+                    epitope_indices=epitope_indices,
+                    pdb_id=pdb_id,
+                    chain_id=chain_id,
+                    num_rbf=self.num_rbf,
+                    num_posenc=self.num_posenc,
+                    radius=self.radius,
+                    verbose=self.verbose
+                )
+                
+                if graph_data is not None:
+                    self.data_list.append(graph_data)
+                else:
+                    failed_proteins.append(f"{pdb_id}_{chain_id}")
+                        
+            except Exception as e:
+                failed_proteins.append(f"{pdb_id}_{chain_id}")
+                if self.verbose:
+                    print(f"Error processing {pdb_id}_{chain_id}: {str(e)}")
+        
+        if failed_proteins and self.verbose:
+            print(f"Failed to process {len(failed_proteins)} proteins: {failed_proteins[:5]}...")
+        
+        # Apply undersampling if specified
+        if self.undersample is not None:
+            self.data_list = apply_undersample(
+                self.data_list,
+                self.undersample,
+                seed=self.seed,
+                verbose=self.verbose
+            )
+        
+        if self.verbose:
+            print(f"Successfully created {len(self.data_list)} protein graphs")
+    
+    def _save_cache(self):
+        """Save processed dataset to cache."""
+        try:
+            self._save_cache_hdf5()
+            if self.verbose:
+                print(f"Dataset cached to {self.cache_file}")
+        except Exception as e:
+            if self.verbose:
+                print(f"Failed to save cache: {str(e)}")
+    
+    def _load_cache(self):
+        """Load processed dataset from cache."""
+        try:
+            self._load_cache_hdf5()
+            if self.verbose:
+                print(f"Loaded {len(self.data_list)} samples from cache")
+        except Exception as e:
+            if self.verbose:
+                print(f"Failed to load cache: {str(e)}")
+            self.data_list = []
+    
+    def _save_cache_hdf5(self):
+        """Save dataset using HDF5 format."""
+        with h5py.File(self.cache_file, 'w') as f:
+            # Save metadata
+            f.attrs['num_samples'] = len(self.data_list)
+            f.attrs['radius'] = self.radius
+            f.attrs['threshold'] = self.threshold
+            f.attrs['data_split'] = self.data_split
+            f.attrs['encoder'] = self.encoder
+            f.attrs['dataset_type'] = 'antigen_full'
+            
+            # Save each protein as a separate group
+            for i, data in enumerate(tqdm(self.data_list, desc="Saving dataset...", disable=not self.verbose)):
+                group = f.create_group(f'protein_{i}')
+                
+                # Save tensors as datasets with compression
+                group.create_dataset('x', data=data.x.numpy(), compression='gzip', compression_opts=6)
+                group.create_dataset('pos', data=data.pos.numpy(), compression='gzip', compression_opts=6)
+                group.create_dataset('rsa', data=data.rsa.numpy(), compression='gzip', compression_opts=6)
+                group.create_dataset('edge_index', data=data.edge_index.numpy(), compression='gzip', compression_opts=6)
+                group.create_dataset('edge_attr', data=data.edge_attr.numpy(), compression='gzip', compression_opts=6)
+                group.create_dataset('y_node', data=data.y_node.numpy(), compression='gzip', compression_opts=6)
+                
+                # Save scalar and list attributes
+                group.attrs['pdb_id'] = data.pdb_id.encode('utf-8')
+                group.attrs['chain_id'] = data.chain_id.encode('utf-8')
+                group.attrs['num_nodes'] = data.num_nodes
+                group.attrs['num_epitopes'] = data.num_epitopes
+                group.attrs['epitope_ratio'] = data.epitope_ratio
+                group.attrs['radius'] = data.radius
+                
+                # Save epitope indices
+                group.create_dataset('epitope_indices', data=np.array(data.epitope_indices), compression='gzip', compression_opts=6)
+    
+    def _load_cache_hdf5(self):
+        """Load dataset from HDF5 cache."""
+        self.data_list = []
+        
+        with h5py.File(self.cache_file, 'r') as f:
+            total_samples = f.attrs['num_samples']
+            
+            for i in tqdm(range(total_samples), desc="Loading dataset...", disable=not self.verbose):
+                group = f[f'protein_{i}']
+                attrs = dict(group.attrs)
+                
+                # Safe string decoding
+                def safe_decode(attr):
+                    val = attrs[attr]
+                    return val.decode('utf-8') if isinstance(val, bytes) else str(val)
+                
+                data = Data(
+                    x=torch.tensor(group['x'][:]),
+                    pos=torch.tensor(group['pos'][:]),
+                    rsa=torch.tensor(group['rsa'][:]),
+                    edge_index=torch.tensor(group['edge_index'][:]),
+                    edge_attr=torch.tensor(group['edge_attr'][:]),
+                    y_node=torch.tensor(group['y_node'][:]),
+                    epitope_indices=group['epitope_indices'][:].tolist(),
+                    pdb_id=safe_decode('pdb_id'),
+                    chain_id=safe_decode('chain_id'),
+                    num_nodes=int(attrs['num_nodes']),
+                    num_epitopes=int(attrs['num_epitopes']),
+                    epitope_ratio=float(attrs['epitope_ratio']),
+                    radius=float(attrs['radius'])
+                )
+                self.data_list.append(data)
+        
+        # Apply undersampling if specified
+        if self.undersample is not None:
+            self.data_list = apply_undersample(
+                self.data_list,
+                self.undersample,
+                seed=self.seed,
+                verbose=self.verbose
+            )
+    
+    def len(self) -> int:
+        """Return the number of samples in the dataset."""
+        return len(self.data_list)
+    
+    def get(self, idx: int) -> Data:
+        """Get a sample by index."""
+        return self.data_list[idx]
+    
+    def get_stats(self) -> Dict:
+        """Get dataset statistics."""
+        if not self.data_list:
+            return {}
+        
+        # Collect statistics
+        num_nodes_list = [data.num_nodes for data in self.data_list]
+        num_edges_list = [data.edge_index.shape[1] for data in self.data_list]
+        num_epitopes_list = [data.num_epitopes for data in self.data_list]
+        epitope_ratio_list = [data.epitope_ratio for data in self.data_list]
+        
+        # Overall statistics
+        total_nodes = sum(num_nodes_list)
+        total_edges = sum(num_edges_list)
+        total_epitopes = sum(num_epitopes_list)
+        
+        stats = {
+            'num_proteins': len(self.data_list),
+            'avg_nodes_per_protein': np.mean(num_nodes_list),
+            'std_nodes_per_protein': np.std(num_nodes_list),
+            'min_nodes_per_protein': np.min(num_nodes_list),
+            'max_nodes_per_protein': np.max(num_nodes_list),
+            'avg_edges_per_protein': np.mean(num_edges_list),
+            'std_edges_per_protein': np.std(num_edges_list),
+            'total_nodes': total_nodes,
+            'total_edges': total_edges,
+            'total_epitopes': total_epitopes,
+            'avg_epitopes_per_protein': np.mean(num_epitopes_list),
+            'avg_epitope_ratio': np.mean(epitope_ratio_list),
+            'overall_epitope_ratio': total_epitopes / total_nodes if total_nodes > 0 else 0,
+        }
+        
+        return stats
+    
+    def print_stats(self):
+        """Print dataset statistics."""
+        stats = self.get_stats()
+        if not stats:
+            print("No statistics available (empty dataset)")
+            return
+        
+        print(f"\n=== {self.data_split.upper()} Antigen Dataset Statistics ===")
+        print(f"Number of proteins: {stats['num_proteins']:,}")
+        print(f"Average nodes per protein: {stats['avg_nodes_per_protein']:.1f} ± {stats['std_nodes_per_protein']:.1f}")
+        print(f"Nodes per protein range: [{stats['min_nodes_per_protein']}, {stats['max_nodes_per_protein']}]")
+        print(f"Average edges per protein: {stats['avg_edges_per_protein']:.1f} ± {stats['std_edges_per_protein']:.1f}")
+        print(f"Total nodes: {stats['total_nodes']:,}")
+        print(f"Total edges: {stats['total_edges']:,}")
+        print(f"Total epitope nodes: {stats['total_epitopes']:,}")
+        print(f"Average epitopes per protein: {stats['avg_epitopes_per_protein']:.1f}")
+        print(f"Average epitope ratio per protein: {stats['avg_epitope_ratio']:.3f}")
+        print(f"Overall epitope ratio: {stats['overall_epitope_ratio']:.3f}")
+        print("=" * 50)
+
+
+class SphereGraphDataset(Dataset):
+    """
+    Optimized graph dataset for training ReGEP model using spherical regions from antigen chains.
+    Each graph represents a spherical region centered on a surface residue.
+    
+    Optimizations:
+    - Only uses HDF5 format for caching
+    - Builds complete dataset without zero_ratio filtering
+    - Applies zero_ratio and undersample during loading
+    - Faster caching with optimized HDF5 structure
+    """
+    
+    def __init__(
+        self,
+        data_split: str = "train",
+        radius: int = 18,
+        threshold: float = 0.25,
+        num_posenc: int = 16,
+        num_rbf: int = 16,
+        zero_ratio: float = 0.1,
+        undersample: Union[int, float, None] = None,
+        cache_dir: Optional[str] = None,
+        force_rebuild: bool = False,
+        verbose: bool = True,
+        seed: int = 42,
+        use_embeddings2: bool = False
+    ):
+        """
+        Initialize the spherical graph dataset.
+        
+        Args:
+            data_split: Data split name ('train', 'val', 'test')
+            radius: Radius for spherical regions
+            threshold: SASA threshold for surface residues
+            num_posenc: Number of positional encoding features
+            num_rbf: Number of RBF features
+            zero_ratio: Ratio to downsample graphs with recall=0 (0.3 means keep 30%)
+            undersample: Undersample parameter (int for count, float for ratio)
+            cache_dir: Directory to cache processed data
+            force_rebuild: Whether to force rebuild the dataset
+            verbose: Whether to print progress information
+            seed: Random seed for reproducibility
+        """
+        self.data_split = data_split
+        self.radius = radius
+        self.threshold = threshold
+        self.num_posenc = num_posenc
+        self.num_rbf = num_rbf
+        self.zero_ratio = zero_ratio
+        self.undersample = undersample
+        self.verbose = verbose
+        self.seed = seed
+        self.use_embeddings2 = use_embeddings2
+        
+        # Set cache directory to large disk
+        if cache_dir is None:
+            cache_dir = Path(f"{BASE_DIR}/data/region_cache/sphere_r{radius}")
+        self.cache_dir = Path(cache_dir)
+        self.cache_dir.mkdir(parents=True, exist_ok=True)
+        
+        # Cache file for this configuration (only HDF5)
+        self.cache_file = self.cache_dir / f"{data_split}_dataset_complete.h5"
+        
+        # Load data splits
+        self.antigens = load_data_split(data_split, verbose=verbose)
+        
+        # Initialize data list
+        self.data_list = []
+        
+        # Load or build dataset
+        if self.cache_file.exists() and not force_rebuild:
+            if verbose:
+                print(f"Loading cached dataset with radius {self.radius} from {self.cache_file}")
+            self._load_cache()
+        else:
+            if verbose:
+                print(f"Building complete dataset with radius {self.radius} for {data_split} split...")
+            self._build_dataset()
+            self._save_cache()
+        
+        super().__init__()
+    
+    def _load_protein_data(self, pdb_id: str, chain_id: str) -> Optional[Dict]:
+        """
+        Load precomputed protein data from files.
+        
+        Args:
+            pdb_id: PDB ID
+            chain_id: Chain ID
+            
+        Returns:
+            Dictionary containing all protein data or None if loading fails
+        """
+        try:
+            protein_key = f"{pdb_id}_{chain_id}"
+            
+            # Load embeddings
+            embedding_file = Path(BASE_DIR) / "data" / "embeddings" / 'esmc' / f"{protein_key}.h5"
+            if not embedding_file.exists():
+                if self.verbose:
+                    print(f"Embedding file not found: {embedding_file}")
+                return None
+            
+            with h5py.File(embedding_file, "r") as h5f:
+                embeddings = h5f["embedding"][:]
+            
+            # Load other embeddings if available
+            esm2_file = Path(BASE_DIR) / "data" / "embeddings" / "esm2" / f"{protein_key}.h5"
+            if not esm2_file.exists():
+                if self.verbose:
+                    print(f"ESM2 file not found: {esm2_file}")
+                embeddings2 = None
+            else:
+                with h5py.File(esm2_file, "r") as h5f:
+                    embeddings2 = h5f["embedding"][:]
+            
+            # Load backbone atoms
+            coords_file = Path(BASE_DIR) / "data" / "coords" / f"{protein_key}.npy"
+            if not coords_file.exists():
+                if self.verbose:
+                    print(f"Coords file not found: {coords_file}")
+                return None
+            backbone_atoms = np.load(coords_file)
+            
+            # Load RSA values
+            rsa_file = Path(BASE_DIR) / "data" / "rsa" / f"{protein_key}.npy"
+            if not rsa_file.exists():
+                if self.verbose:
+                    print(f"RSA file not found: {rsa_file}")
+                return None
+            rsa_values = np.load(rsa_file)
+            
+            # Load surface coverage data
+            sphere_file = Path(BASE_DIR) / "data" / "antigen_sphere" / f"{protein_key}.h5"
+            radius_key = f"r{self.radius}"
+            
+            if not sphere_file.exists():
+                if self.verbose:
+                    print(f"Sphere file not found: {sphere_file}")
+                return None
+            
+            coverage_dict = {}
+            with h5py.File(sphere_file, "r") as h5f:
+                if radius_key not in h5f:
+                    if self.verbose:
+                        print(f"Radius {self.radius} not found in {sphere_file}")
+                    return None
+                
+                radius_group = h5f[radius_key]
+                for center_idx_str in radius_group.keys():
+                    center_idx = int(center_idx_str)
+                    center_group = radius_group[center_idx_str]
+                    covered_indices = center_group['covered_indices'][:].tolist()
+                    covered_epitope_indices = center_group['covered_epitope_indices'][:].tolist()
+                    precision = float(center_group.attrs['precision'])
+                    recall = float(center_group.attrs['recall'])
+                    coverage_dict[center_idx] = (covered_indices, covered_epitope_indices, precision, recall)
+            
+            # Load epitope data
+            _, _, epitopes = load_epitopes_csv()
+            binary_labels = epitopes.get(protein_key, [])
+            
+            # Create epitope indices
+            epitope_indices = []
+            for idx, is_epitope in enumerate(binary_labels):
+                if is_epitope == 1:
+                    epitope_indices.append(idx)
+            
+            return {
+                'embeddings': embeddings,
+                'backbone_atoms': backbone_atoms,
+                'rsa_values': rsa_values,
+                'coverage_dict': coverage_dict,
+                'epitope_indices': epitope_indices,
+                'embeddings2': embeddings2
+            }
+            
+        except Exception as e:
+            if self.verbose:
+                print(f"Error loading protein data for {pdb_id}_{chain_id}: {str(e)}")
+            return None
+
+    def _build_dataset(self):
+        """Build the complete dataset from precomputed data files (no zero_ratio filtering)."""
+        failed_proteins = []
+        
+        for pdb_id, chain_id in tqdm(self.antigens, desc=f"Processing {self.data_split} antigens", 
+                                   disable=not self.verbose):
+            try:
+                # Load precomputed data directly
+                protein_data = self._load_protein_data(pdb_id, chain_id)
+                if protein_data is None:
+                    if self.verbose:
+                        print(f"Failed to load data for {pdb_id}_{chain_id}")
+                    continue
+                
+                embeddings = protein_data['embeddings']
+                embeddings2 = protein_data['embeddings2']
+                backbone_atoms = protein_data['backbone_atoms']
+                rsa_values = protein_data['rsa_values']
+                coverage_dict = protein_data['coverage_dict']
+                epitope_indices = protein_data['epitope_indices']
+                
+                if not coverage_dict:
+                    if self.verbose:
+                        print(f"No surface regions found for {pdb_id}_{chain_id}")
+                    continue
+                
+                # Process each spherical region (no zero_ratio filtering here)
+                for center_idx, (covered_indices, covered_epitope_indices, precision, recall) in coverage_dict.items():
+                    if len(covered_indices) < 2:  # Skip regions with too few residues
+                        continue
+                    
+                    # Create graph data for this region (include all data)
+                    graph_data = create_graph_data(
+                        center_idx=center_idx,
+                        covered_indices=covered_indices,
+                        covered_epitope_indices=covered_epitope_indices,
+                        embeddings=embeddings,
+                        embeddings2=embeddings2,
+                        backbone_atoms=backbone_atoms,
+                        rsa_values=rsa_values,
+                        epitope_indices=epitope_indices,
+                        recall=recall,
+                        precision=precision,
+                        pdb_id=pdb_id,
+                        chain_id=chain_id,
+                        num_rbf=self.num_rbf,
+                        num_posenc=self.num_posenc,
+                        verbose=self.verbose
+                    )
+                    
+                    if graph_data is not None:
+                        self.data_list.append(graph_data)
+                        
+            except Exception as e:
+                failed_proteins.append(f"{pdb_id}_{chain_id}")
+                if self.verbose:
+                    print(f"Error processing {pdb_id}_{chain_id}: {str(e)}")
+        
+        if failed_proteins and self.verbose:
+            print(f"Failed to process {len(failed_proteins)} proteins: {failed_proteins[:5]}...")
+        
+        if self.verbose:
+            print(f"Successfully created {len(self.data_list)} graph samples (complete dataset)")
+    
+    def _save_cache(self):
+        """Save processed dataset to cache."""
+        try:
+            self._save_cache_hdf5()
+            if self.verbose:
+                print(f"Dataset cached to {self.cache_file}")
+        except Exception as e:
+            if self.verbose:
+                print(f"Failed to save cache: {str(e)}")
+    
+    def _load_cache(self):
+        """Load processed dataset from cache."""
+        try:
+            self._load_cache_hdf5()
+            if self.verbose:
+                print(f"Loaded {len(self.data_list)} samples from cache")
+        except Exception as e:
+            if self.verbose:
+                print(f"Failed to load cache: {str(e)}")
+            self.data_list = []
+    
+    def _save_cache_hdf5(self):
+        """Save dataset using optimized HDF5 format for faster loading."""
+        with h5py.File(self.cache_file, 'w') as f:
+            # Save metadata
+            f.attrs['num_samples'] = len(self.data_list)
+            f.attrs['radius'] = self.radius
+            f.attrs['threshold'] = self.threshold
+            f.attrs['data_split'] = self.data_split
+            f.attrs['complete_dataset'] = True  # Mark as complete dataset
+            
+            # Pre-allocate arrays for better performance
+            num_samples = len(self.data_list)
+            if num_samples == 0:
+                return
+            
+            # Collect all data first to determine max dimensions
+            all_x = []
+            all_pos = []
+            all_rsa = []
+            all_edge_index = []
+            all_edge_attr = []
+            all_y = []
+            all_y_node = []
+            all_center_idx = []
+            all_precision = []
+            all_pdb_ids = []
+            all_chain_ids = []
+            all_num_nodes = []
+            all_covered_indices = []
+            all_embeddings2 = []
+            
+            max_nodes = 0
+            max_edges = 0
+            
+            for data in self.data_list:
+                all_x.append(data.x.numpy())
+                all_pos.append(data.pos.numpy())
+                all_rsa.append(data.rsa.numpy())
+                all_edge_index.append(data.edge_index.numpy())
+                all_edge_attr.append(data.edge_attr.numpy())
+                all_y.append(data.y.numpy())
+                all_y_node.append(data.y_node.numpy())
+                all_center_idx.append(data.center_idx)
+                all_precision.append(data.precision)
+                all_pdb_ids.append(data.pdb_id.encode('utf-8'))
+                all_chain_ids.append(data.chain_id.encode('utf-8'))
+                all_num_nodes.append(data.num_nodes)
+                all_covered_indices.append(data.covered_indices)
+                
+                # Handle embeddings2 safely - it could be None or numpy array
+                if hasattr(data, 'embeddings2') and data.embeddings2 is not None:
+                    if isinstance(data.embeddings2, np.ndarray):
+                        all_embeddings2.append(data.embeddings2)
+                    else:
+                        # It's a torch tensor
+                        all_embeddings2.append(data.embeddings2.numpy())
+                else:
+                    # No embeddings2 available, use zeros as placeholder
+                    all_embeddings2.append(np.zeros((data.num_nodes, 1280), dtype=np.float32))  # ESM2 dim
+                
+                max_nodes = max(max_nodes, data.num_nodes)
+                max_edges = max(max_edges, data.edge_index.shape[1])
+            
+            # Save each graph as a separate group with compression
+            progress_bar = tqdm(enumerate(self.data_list), total=num_samples, desc="Saving dataset...", disable=not self.verbose)
+            
+            for i, data in progress_bar:
+                group = f.create_group(f'graph_{i}')
+                
+                # Save tensors as datasets with compression
+                group.create_dataset('x', data=all_x[i], compression='gzip', compression_opts=6)
+                group.create_dataset('pos', data=all_pos[i], compression='gzip', compression_opts=6)
+                group.create_dataset('rsa', data=all_rsa[i], compression='gzip', compression_opts=6)
+                group.create_dataset('edge_index', data=all_edge_index[i], compression='gzip', compression_opts=6)
+                group.create_dataset('edge_attr', data=all_edge_attr[i], compression='gzip', compression_opts=6)
+                group.create_dataset('y', data=all_y[i], compression='gzip', compression_opts=6)
+                group.create_dataset('y_node', data=all_y_node[i], compression='gzip', compression_opts=6)
+                group.create_dataset('embeddings2', data=all_embeddings2[i], compression='gzip', compression_opts=6)
+                
+                # Save scalar attributes
+                group.attrs['center_idx'] = all_center_idx[i]
+                group.attrs['precision'] = all_precision[i]
+                group.attrs['pdb_id'] = all_pdb_ids[i]
+                group.attrs['chain_id'] = all_chain_ids[i]
+                group.attrs['num_nodes'] = all_num_nodes[i]
+                
+                # Save list attributes as datasets with compression
+                group.create_dataset('covered_indices', data=np.array(all_covered_indices[i]), compression='gzip', compression_opts=6)
+                
+    def _load_cache_hdf5(self):
+        """Optimized cache loader with robust string handling."""
+        self.data_list = []
+        
+        with h5py.File(self.cache_file, 'r') as f:
+            # PHASE 1: Rapid metadata scan
+            zero_recall_indices = []
+            non_zero_recall_indices = []
+            total_samples = f.attrs['num_samples']
+            
+            if self.verbose:
+                print(f"Scanning {total_samples} samples for recall values...")
+            
+            for i in range(total_samples):
+                recall = f[f'graph_{i}/y'][0].item()
+                if recall == 0.0:
+                    zero_recall_indices.append(i)
+                else:
+                    non_zero_recall_indices.append(i)
+            
+            # PHASE 2: Apply zero_ratio filtering
+            selected_indices = non_zero_recall_indices.copy()
+            
+            if isinstance(self.zero_ratio, (int, float)) and 0 <= self.zero_ratio <= 1:
+                if self.zero_ratio < 1.0 and zero_recall_indices:
+                    random.seed(self.seed)
+                    target_count = int(len(zero_recall_indices) * self.zero_ratio)
+                    selected_zero_indices = random.sample(zero_recall_indices, target_count)
+                    selected_indices.extend(selected_zero_indices)
+                    
+                    if self.verbose:
+                        kept = len(selected_zero_indices)
+                        total = len(zero_recall_indices)
+                        print(f"Zero-recall filtering: kept {kept}/{total} samples (ratio={self.zero_ratio})")
+                else:
+                    selected_indices.extend(zero_recall_indices)
+            
+            # PHASE 3: Selective data loading with safe string handling
+            if self.verbose:
+                print(f"Loading {len(selected_indices)} selected samples...")
+            
+            for idx in tqdm(selected_indices, disable=not self.verbose):
+                group = f[f'graph_{idx}']
+                attrs = dict(group.attrs)
+                
+                # Safe string decoding
+                def safe_decode(attr):
+                    val = attrs[attr]
+                    return val.decode('utf-8') if isinstance(val, bytes) else str(val)
+                
+                # Load embeddings2 if available and use_embeddings2 is True
+                if 'embeddings2' in group and self.use_embeddings2:
+                    if group['embeddings2'] is not None:
+                        emb = torch.tensor(group['embeddings2'][:])
+                    else:
+                        emb = torch.tensor(group['x'][:])
+                else:
+                    emb = torch.tensor(group['x'][:])
+                
+                data = Data(
+                    x=emb,
+                    pos=torch.tensor(group['pos'][:]),
+                    rsa=torch.tensor(group['rsa'][:]),
+                    edge_index=torch.tensor(group['edge_index'][:]),
+                    edge_attr=torch.tensor(group['edge_attr'][:]),
+                    y=torch.tensor(group['y'][:]),
+                    y_node=torch.tensor(group['y_node'][:]),
+                    center_idx=int(attrs['center_idx']),
+                    covered_indices=group['covered_indices'][:].tolist(),
+                    precision=float(attrs['precision']),
+                    pdb_id=safe_decode('pdb_id'),
+                    chain_id=safe_decode('chain_id'),
+                    num_nodes=int(attrs['num_nodes'])
+                )
+                
+                self.data_list.append(data)
+        
+        # PHASE 4: Apply undersampling
+        if self.undersample is not None:
+            self.data_list = apply_undersample(
+                self.data_list,
+                self.undersample,
+                seed=self.seed,
+                verbose=self.verbose
+            )
+        
+        if self.verbose:
+            print(f"Loaded {len(self.data_list)} samples (optimized loader)")
+    
+    def len(self) -> int:
+        """Return the number of samples in the dataset."""
+        return len(self.data_list)
+    
+    def get(self, idx: int) -> Data:
+        """Get a sample by index."""
+        return self.data_list[idx]
+    
+    def apply_filters(self, zero_ratio: Optional[float] = None, undersample: Union[int, float, None] = None, seed: int = None):
+        """
+        Apply filtering to the already loaded dataset (for compatibility).
+        Note: It's more efficient to set these parameters during initialization.
+        
+        Args:
+            zero_ratio: Ratio to downsample graphs with recall=0
+            undersample: Undersample parameter
+            seed: Random seed for reproducibility
+        """
+        if seed is None:
+            seed = self.seed
+            
+        # Update instance parameters and re-filter
+        if zero_ratio is not None:
+            self.zero_ratio = zero_ratio
+        if undersample is not None:
+            self.undersample = undersample
+        if seed is not None:
+            self.seed = seed
+            
+        # Reload from cache with new parameters
+        if self.cache_file.exists():
+            if self.verbose:
+                print("Re-applying filters to cached dataset...")
+            self._load_cache_hdf5()
+        else:
+            if self.verbose:
+                print("Warning: No cache file found, filters cannot be applied")
+    
+    def get_stats(self) -> Dict:
+        """Get dataset statistics."""
+        if not self.data_list:
+            return {}
+        
+        # Collect statistics
+        num_nodes_list = [data.num_nodes for data in self.data_list]
+        recall_list = [data.y.item() for data in self.data_list]
+        precision_list = [data.precision for data in self.data_list]
+        
+        # Node-level statistics
+        total_nodes = sum(num_nodes_list)
+        total_epitopes = sum([data.y_node.sum().item() for data in self.data_list])
+        num_zero_recall = sum([1 for data in self.data_list if data.y.item() == 0])
+        
+        stats = {
+            'num_graphs': len(self.data_list),
+            'avg_nodes_per_graph': np.mean(num_nodes_list),
+            'std_nodes_per_graph': np.std(num_nodes_list),
+            'min_nodes_per_graph': np.min(num_nodes_list),
+            'max_nodes_per_graph': np.max(num_nodes_list),
+            'total_nodes': total_nodes,
+            'total_epitopes': total_epitopes,
+            'epitope_ratio': total_epitopes / total_nodes if total_nodes > 0 else 0,
+            'avg_recall': np.mean(recall_list),
+            'std_recall': np.std(recall_list),
+            'avg_precision': np.mean(precision_list),
+            'std_precision': np.std(precision_list),
+            'num_zero_recall': num_zero_recall,
+        }
+        
+        return stats
+    
+    def print_stats(self):
+        """Print dataset statistics."""
+        stats = self.get_stats()
+        if not stats:
+            print("No statistics available (empty dataset)")
+            return
+        
+        print(f"\n=== {self.data_split.upper()} Dataset Statistics ===")
+        print(f"Number of graphs: {stats['num_graphs']:,}")
+        print(f"Average nodes per graph: {stats['avg_nodes_per_graph']:.1f} ± {stats['std_nodes_per_graph']:.1f}")
+        print(f"Nodes per graph range: [{stats['min_nodes_per_graph']}, {stats['max_nodes_per_graph']}]")
+        print(f"Total nodes: {stats['total_nodes']:,}")
+        print(f"Total epitope nodes: {stats['total_epitopes']:,}")
+        print(f"Epitope ratio: {stats['epitope_ratio']:.3f}")
+        print(f"Average recall: {stats['avg_recall']:.3f} ± {stats['std_recall']:.3f}")
+        print(f"Average precision: {stats['avg_precision']:.3f} ± {stats['std_precision']:.3f}")
+        print(f"Number of graphs with zero recall: {stats['num_zero_recall']:,}")
+        print("=" * 40)
+
+
+class MultiRadiusGraphDataset(Dataset):
+    """
+    Dataset that combines multiple radius datasets for multi-scale training.
+    """
+    
+    def __init__(
+        self,
+        data_split: str = "train",
+        radii: List[int] = [16, 18, 20],
+        threshold: float = 0.25,
+        num_posenc: int = 16,
+        num_rbf: int = 16,
+        zero_ratio: float = 0.1,
+        undersample: Union[int, float, None] = None,
+        cache_dir: Optional[str] = None,
+        force_rebuild: bool = False,
+        verbose: bool = True,
+        use_embeddings2: bool = False
+    ):
+        """
+        Initialize multi-radius dataset.
+        
+        Args:
+            data_split: Data split name
+            radii: List of radii to use
+            threshold: SASA threshold for surface residues
+            num_posenc: Number of positional encoding features
+            num_rbf: Number of RBF features
+            zero_ratio: Ratio to downsample graphs with recall=0
+            undersample: Undersample parameter (int for count, float for ratio)
+            cache_dir: Directory to cache processed data
+            force_rebuild: Whether to force rebuild the dataset
+            verbose: Whether to print progress information
+        """
+        self.data_split = data_split
+        self.radii = radii
+        self.verbose = verbose
+        
+        # Create individual datasets
+        self.datasets = []
+        for radius in radii:
+            dataset = SphereGraphDataset(
+                data_split=data_split,
+                radius=radius,
+                threshold=threshold,
+                num_posenc=num_posenc,
+                num_rbf=num_rbf,
+                zero_ratio=zero_ratio,
+                undersample=undersample,
+                cache_dir=cache_dir,
+                force_rebuild=force_rebuild,
+                verbose=verbose,
+                use_embeddings2=use_embeddings2
+            )
+            self.datasets.append(dataset)
+        
+        # Combine all data
+        self.data_list = []
+        for dataset in self.datasets:
+            self.data_list.extend(dataset.data_list)
+        
+        if verbose:
+            print(f"Combined {len(self.datasets)} datasets with {len(self.data_list)} total samples")
+        
+        super().__init__()
+    
+    def len(self) -> int:
+        return len(self.data_list)
+    
+    def get(self, idx: int) -> Data:
+        return self.data_list[idx]
+    
+    def apply_filters(self, undersample: Union[int, float, None] = None, seed: int = 42):
+        """
+        Apply filtering to the loaded multi-radius dataset.
+        
+        Args:
+            undersample: Undersample parameter (int for count, float for ratio)
+            seed: Random seed for reproducibility
+        """
+        if undersample is not None:
+            original_size = len(self.data_list)
+            self.data_list = apply_undersample(self.data_list, undersample, seed=seed, verbose=True)
+    
+    def get_stats(self) -> Dict:
+        """Get combined dataset statistics."""
+        if not self.data_list:
+            return {}
+        
+        # Collect statistics
+        num_nodes_list = [data.num_nodes for data in self.data_list]
+        recall_list = [data.y.item() for data in self.data_list]
+        
+        # Node-level statistics
+        total_nodes = sum(num_nodes_list)
+        total_epitopes = sum([data.y_node.sum().item() for data in self.data_list])
+        
+        stats = {
+            'num_graphs': len(self.data_list),
+            'num_radii': len(self.radii),
+            'radii': self.radii,
+            'avg_nodes_per_graph': np.mean(num_nodes_list),
+            'std_nodes_per_graph': np.std(num_nodes_list),
+            'min_nodes_per_graph': np.min(num_nodes_list),
+            'max_nodes_per_graph': np.max(num_nodes_list),
+            'total_nodes': total_nodes,
+            'total_epitopes': total_epitopes,
+            'epitope_ratio': total_epitopes / total_nodes if total_nodes > 0 else 0,
+            'avg_recall': np.mean(recall_list),
+        }
+        
+        return stats
+
+    def print_stats(self):
+        """Print dataset statistics."""
+        stats = self.get_stats()
+        if not stats:
+            print("No statistics available (empty dataset)")
+            return
+        
+        print(f"\n=== {self.data_split.upper()} Dataset Statistics ===")
+        print(f"Number of graphs: {stats['num_graphs']:,}")
+        print(f"Average nodes per graph: {stats['avg_nodes_per_graph']:.1f} ± {stats['std_nodes_per_graph']:.1f}")
+        print(f"Nodes per graph range: [{stats['min_nodes_per_graph']}, {stats['max_nodes_per_graph']}]")
+        print(f"Total nodes: {stats['total_nodes']:,}")
+        print(f"Total epitope nodes: {stats['total_epitopes']:,}")
+        print(f"Epitope ratio: {stats['epitope_ratio']:.3f}")
+        print(f"Average recall: {stats['avg_recall']:.3f} ± {stats['std_recall']:.3f}")
+        print(f"Average precision: {stats['avg_precision']:.3f} ± {stats['std_precision']:.3f}")
+        print("=" * 40)
+
+
+
+# Utility functions for dataset creation and management
+def create_datasets(
+    radii: List[int] = [16, 18, 20],
+    splits: List[str] = ["train", "test"],
+    threshold: float = 0.25,
+    zero_ratio: float = None,
+    undersample: Union[int, float, None] = None,
+    cache_dir: Optional[str] = None,
+    force_rebuild: bool = False,
+    verbose: bool = False,
+    seed: int = 42,
+    use_embeddings2: bool = False,
+) -> Dict[str, SphereGraphDataset]:
+    """
+    Create optimized datasets for all splits and radii.
+    
+    Args:
+        radii: List of radii to use
+        splits: List of data splits to create
+        threshold: SASA threshold for surface residues
+        zero_ratio: Ratio to downsample graphs with recall=0
+        undersample: Undersample parameter (int for count, float for ratio)
+        cache_dir: Directory to cache processed data
+        force_rebuild: Whether to force rebuild datasets
+        verbose: Whether to print progress information
+        seed: Random seed for reproducibility
+        
+    Returns:
+        Dictionary mapping split names to datasets
+    """
+    datasets = {}
+    
+    for split in splits:
+        if len(radii) == 1:
+            # Single radius dataset
+            dataset = SphereGraphDataset(
+                data_split=split,
+                radius=radii[0],
+                threshold=threshold,
+                zero_ratio=zero_ratio,
+                undersample=undersample,
+                cache_dir=cache_dir,
+                force_rebuild=force_rebuild,
+                verbose=verbose,
+                seed=seed,
+                use_embeddings2=use_embeddings2
+            )
+            if verbose:
+                dataset.print_stats()
+        else:
+            # Multi-radius dataset
+            dataset = MultiRadiusGraphDataset(
+                data_split=split,
+                radii=radii,
+                threshold=threshold,
+                zero_ratio=zero_ratio,
+                undersample=undersample,
+                cache_dir=cache_dir,
+                force_rebuild=force_rebuild,
+                verbose=verbose,
+                use_embeddings2=use_embeddings2
+            )
+        
+        datasets[split] = dataset
+    
+    return datasets
+
+
+def custom_collate_fn(batch):
+    """
+    Custom collate function for ReGEP model.
+    Converts PyG Data objects to the format expected by ReGEP.
+    """
+    # Use PyG's default batching
+    batched_data = Batch.from_data_list(batch)
+    
+    # ReGEP expects the input features to be concatenated
+    # x: [N_total, embed_dim], rsa: [N_total], ss: [N_total, 2]
+    # The model will concatenate them internally: [x, rsa, ss] -> [N_total, embed_dim + 3]
+    
+    return batched_data
+
+
+class ReGEPDataLoader(DataLoader):
+    """
+    Custom DataLoader for ReGEP model that handles the specific input format.
+    Supports undersampling at the DataLoader level.
+    """
+    
+    def __init__(self, dataset, batch_size=32, shuffle=True, **kwargs):
+        """
+        Initialize ReGEP DataLoader with optional undersampling.
+        
+        Args:
+            dataset: The dataset to load from
+            batch_size: Batch size
+            shuffle: Whether to shuffle the data
+            **kwargs: Additional arguments for DataLoader
+        """
+        # Set default collate_fn if not provided
+        if 'collate_fn' not in kwargs:
+            kwargs['collate_fn'] = custom_collate_fn
+            
+        super().__init__(
+            dataset=dataset,
+            batch_size=batch_size,
+            shuffle=shuffle,
+            **kwargs
+        )
+
+def create_data_loader(
+    radii=[16, 18, 20],
+    batch_size=32,
+    zero_ratio=0.1,
+    undersample=0.5,
+    seed=42,
+    verbose=False,
+    use_embeddings2=False,
+    **kwargs
+):
+    """
+    Create train and test data loaders.
+    
+    Args:
+        radii (list): List of radii for data processing
+        batch_size (int): Batch size for training
+        zero_ratio (float): Ratio of zero samples for training
+        undersample (float): Undersampling ratio for training
+        seed (int): Random seed
+        verbose (bool): Whether to print verbose information
+        **kwargs: Additional arguments for data loader
+    
+    Returns:
+        tuple: (train_loader, test_loader)
+    """
+    train_dataset = create_datasets(
+            radii=radii,
+            splits=["train"],
+            threshold=0.25,
+            undersample=undersample,
+            zero_ratio=zero_ratio,
+            cache_dir=None,
+            seed=seed,
+            verbose=verbose,
+            use_embeddings2=use_embeddings2
+        )["train"]
+    
+    test_dataset = create_datasets(
+            radii=radii,
+            splits=["test"],
+            threshold=0.25,
+            undersample=None,
+            zero_ratio=None,
+            cache_dir=None,
+            verbose=verbose,
+            use_embeddings2=use_embeddings2
+        )["test"]
+    
+    train_loader = ReGEPDataLoader(
+        train_dataset,
+        batch_size=batch_size,
+        shuffle=True,
+        collate_fn=custom_collate_fn,
+        **kwargs
+    )
+    
+    test_loader = ReGEPDataLoader(
+        test_dataset,
+        batch_size=batch_size*4,
+        shuffle=False,
+        **kwargs
+    )
+    
+    return train_loader, test_loader
+    

src/bce/data/utils.py

@@ -0,0 +1,352 @@
+"""
+Simplified graph utilities for SphereGraphDataset.
+Contains only the essential functions needed without external dependencies.
+"""
+
+import numpy as np
+from pathlib import Path
+from typing import List, Optional, Tuple
+
+import torch
+from torch_geometric.data import Data
+from torch_geometric.nn import radius_graph
+
+def parse_binding_site_txt(txt_path: Path) -> Tuple[List[str], List[str], List[torch.Tensor]]:
+    """
+    Parse a fasta-like txt file with 3 lines per entry: >id, sequence, label
+    Returns: (rna_ids, rna_seqs, labels)
+    """
+    rna_ids, rna_seqs, labels = [], [], []
+    with open(txt_path, 'r') as f:
+        lines = [line.strip() for line in f if line.strip()]
+    for i in range(0, len(lines), 3):
+        rna_id = lines[i][1:] if lines[i].startswith('>') else lines[i]
+        seq = lines[i+1]
+        label_str = lines[i+2]
+        label = torch.tensor([int(x) for x in label_str], dtype=torch.float32)
+        rna_ids.append(rna_id)
+        rna_seqs.append(seq)
+        labels.append(label)
+    return rna_ids, rna_seqs, labels
+
+def create_graph_data_full(
+    embeddings: np.ndarray,
+    backbone_atoms: np.ndarray,
+    rsa_values: np.ndarray,
+    epitope_indices: List[int],
+    pdb_id: str,
+    chain_id: str,
+    num_rbf: int = 16,
+    num_posenc: int = 16,
+    radius: float = 18.0,
+    verbose: bool = True
+    ) -> Optional[Data]:
+    """
+    Create a PyTorch Geometric Data object for a full protein graph.
+    
+    Args:
+        embeddings: Full protein embeddings [seq_len, embed_dim]
+        backbone_atoms: Full protein backbone atoms [seq_len, 3, 3] (N, CA, C)
+        rsa_values: Full protein RSA values [seq_len]
+        epitope_indices: List of epitope residue indices
+        pdb_id: PDB ID
+        chain_id: Chain ID
+        num_rbf: Number of RBF features
+        num_posenc: Number of positional encoding features
+        radius: Distance threshold for edge creation (default: 18.0 Å)
+        verbose: Whether to print debug information
+        
+    Returns:
+        PyTorch Geometric Data object or None if creation fails
+    """
+    try:
+        # Validate input dimensions
+        seq_len = len(embeddings)
+        if len(backbone_atoms) != seq_len or len(rsa_values) != seq_len:
+            if verbose:
+                print(f"[WARNING] Dimension mismatch for {pdb_id}_{chain_id}: "
+                      f"embeddings={len(embeddings)}, backbone={len(backbone_atoms)}, "
+                      f"rsa={len(rsa_values)}")
+            return None
+        
+        if seq_len == 0:
+            if verbose:
+                print(f"[WARNING] Empty protein {pdb_id}_{chain_id}")
+            return None
+        
+        # Create node labels (binary epitope classification)
+        node_labels = np.zeros(seq_len, dtype=np.float32)
+        if epitope_indices:
+            # Filter epitope_indices to ensure they are within bounds
+            valid_epitope_indices = [idx for idx in epitope_indices if 0 <= idx < seq_len]
+            if valid_epitope_indices:
+                node_labels[valid_epitope_indices] = 1.0
+            
+            if verbose and len(valid_epitope_indices) != len(epitope_indices):
+                print(f"[WARNING] Some epitope indices out of bounds for {pdb_id}_{chain_id}: "
+                      f"filtered {len(epitope_indices)} -> {len(valid_epitope_indices)}")
+        
+        # Extract CA coordinates for distance calculation
+        ca_coords = backbone_atoms[:, 1, :]  # CA is the second atom [seq_len, 3]
+        
+        # Validate CA coordinates
+        if ca_coords.shape[0] == 0:
+            if verbose:
+                print(f"[WARNING] Empty CA coordinates for {pdb_id}_{chain_id}")
+            return None
+        
+        # Check for NaN or infinite values
+        if np.any(np.isnan(ca_coords)) or np.any(np.isinf(ca_coords)):
+            if verbose:
+                print(f"[WARNING] Invalid CA coordinates (NaN/Inf) for {pdb_id}_{chain_id}")
+            return None
+        
+        # Create edges based on distance threshold using radius_graph
+        ca_coords_tensor = torch.tensor(ca_coords, dtype=torch.float32)
+        
+        # Additional safety check for tensor
+        if ca_coords_tensor.numel() == 0:
+            if verbose:
+                print(f"[WARNING] Empty CA coordinates tensor for {pdb_id}_{chain_id}")
+            return None
+        
+        edge_index = radius_graph(ca_coords_tensor, r=radius, loop=False, max_num_neighbors=32)
+        
+        if edge_index.shape[1] == 0:
+            if verbose:
+                print(f"[WARNING] No edges found for {pdb_id}_{chain_id} with radius {radius}")
+            # Create a minimal graph with self-loops to avoid empty graph
+            edge_index = torch.stack([torch.arange(seq_len), torch.arange(seq_len)], dim=0)
+        
+        # Compute edge features
+        edge_features = compute_edge_features(ca_coords, edge_index, num_rbf=num_rbf, num_posenc=num_posenc)
+        
+        # Convert to tensors
+        x = torch.tensor(embeddings, dtype=torch.float32)  # [seq_len, embed_dim]
+        pos = torch.tensor(backbone_atoms, dtype=torch.float32)  # [seq_len, 3, 3]
+        rsa = torch.tensor(rsa_values, dtype=torch.float32)  # [seq_len]
+        
+        # Node-level labels
+        y_node = torch.tensor(node_labels, dtype=torch.float32)  # [seq_len]
+        
+        # Additional protein-level statistics
+        num_epitopes = int(node_labels.sum())
+        epitope_ratio = num_epitopes / seq_len if seq_len > 0 else 0.0
+        
+        # Create Data object
+        data = Data(
+            x=x,  # Node embeddings [seq_len, embed_dim]
+            pos=pos,  # Backbone coordinates [seq_len, 3, 3]
+            rsa=rsa,  # RSA values [seq_len]
+            edge_index=edge_index,  # Edge connectivity [2, n_edges]
+            edge_attr=edge_features,  # Edge features [n_edges, edge_dim]
+            y_node=y_node,  # Node-level labels [seq_len]
+            epitope_indices=epitope_indices,  # Original epitope indices
+            pdb_id=pdb_id,  # PDB ID
+            chain_id=chain_id,  # Chain ID
+            num_nodes=seq_len,  # Number of nodes (residues)
+            num_epitopes=num_epitopes,  # Number of epitope residues
+            epitope_ratio=epitope_ratio,  # Ratio of epitope residues
+            radius=radius  # Distance threshold used for edges
+        )
+        
+        if verbose:
+            print(f"[INFO] Created full protein graph for {pdb_id}_{chain_id}: "
+                  f"{seq_len} nodes, {edge_index.shape[1]} edges, {num_epitopes} epitopes")
+        
+        return data
+        
+    except Exception as e:
+        if verbose:
+            print(f"[ERROR] Failed to create full protein graph for {pdb_id}_{chain_id}: {str(e)}")
+        return None
+
+def create_graph_data(
+    center_idx: int,
+    covered_indices: List[int],
+    covered_epitope_indices: List[int],
+    embeddings: np.ndarray,
+    backbone_atoms: np.ndarray,
+    rsa_values: np.ndarray,
+    epitope_indices: List[int],
+    recall: float,
+    precision: float,
+    pdb_id: str,
+    chain_id: str,
+    embeddings2: np.ndarray = None,
+    num_rbf: int = 16,
+    num_posenc: int = 16,
+    verbose: bool = True
+    ) -> Optional[Data]:
+    """
+    Create a PyTorch Geometric Data object for a spherical region.
+        
+    Args:
+        center_idx: Index of the center residue
+        covered_indices: List of residue indices in the region
+        covered_epitope_indices: List of epitope residue indices in the region
+        embeddings: Full protein embeddings
+        backbone_atoms: Full protein backbone atoms [seq_len, 3, 3]
+        rsa_values: Full protein RSA values
+        epitope_indices: List of all epitope indices in the protein (if available)
+        recall: Region recall value (if available)
+        precision: Region precision value (if available)
+        pdb_id: PDB ID
+        chain_id: Chain ID
+            
+        Returns:
+            PyTorch Geometric Data object or None if creation fails
+        """
+    try:
+        # Validate indices first
+        if not covered_indices:
+            if verbose:
+                print(f"[WARNING] Empty covered_indices for center {center_idx}")
+            return None
+        
+        # Check if indices are within bounds
+        max_idx = max(covered_indices)
+        if max_idx >= len(embeddings) or max_idx >= len(backbone_atoms) or max_idx >= len(rsa_values):
+            if verbose:
+                print(f"[WARNING] Index out of bounds: max_idx={max_idx}, "
+                      f"embeddings_len={len(embeddings)}, backbone_len={len(backbone_atoms)}, "
+                      f"rsa_len={len(rsa_values)}")
+            return None
+        
+        # Extract node features for covered residues
+        node_embeddings = embeddings[covered_indices]  # [n_nodes, embed_dim]
+        node_backbone = backbone_atoms[covered_indices]  # [n_nodes, 3, 3]  
+        node_rsa = rsa_values[covered_indices]  # [n_nodes]
+        
+        if embeddings2 is not None:
+            node_embeddings2 = embeddings2[covered_indices]  # [n_nodes, embed_dim]
+        else:
+            node_embeddings2 = None
+            
+        # Create node labels (binary epitope classification)
+        node_labels = np.zeros(len(covered_indices), dtype=np.float32)
+        # Use the epitope_indices from the loaded data if available
+        epitope_mask = np.isin(covered_indices, epitope_indices)
+        node_labels[epitope_mask] = 1.0
+            
+        # Create fully connected edge index (no self-loops)
+        n_nodes = len(covered_indices)
+        edge_index = get_edges(n_nodes)
+        edge_index = torch.tensor(edge_index, dtype=torch.long)
+        
+        # Compute edge features using CA coordinates
+        ca_coords = node_backbone[:, 1, :]  # Extract CA coordinates [n_nodes, 3]
+        edge_features = compute_edge_features(ca_coords, edge_index, num_rbf=num_rbf, num_posenc=num_posenc)
+        
+        # Convert to tensors
+        x = torch.tensor(node_embeddings, dtype=torch.float32)
+        pos = torch.tensor(node_backbone, dtype=torch.float32)  # [n_nodes, 3, 3]
+        rsa = torch.tensor(node_rsa, dtype=torch.float32)
+        
+        # Graph-level label (recall)
+        y_graph = torch.tensor([recall], dtype=torch.float32)
+        
+        # Node-level labels
+        y_node = torch.tensor(node_labels, dtype=torch.float32)
+        
+        # Create Data object
+        data = Data(
+            x=x,  # Node embeddings [n_nodes, embed_dim]
+            pos=pos,  # Backbone coordinates [n_nodes, 3, 3]
+            rsa=rsa,  # RSA values [n_nodes]
+            edge_index=edge_index,  # Edge connectivity [2, n_edges]
+            edge_attr=edge_features,  # Edge features [n_edges, edge_dim]
+            y=y_graph,  # Graph-level label (recall)
+            y_node=y_node,  # Node-level labels [n_nodes]
+            center_idx=center_idx,  # Center residue index
+            covered_indices=covered_indices,  # All covered residue indices
+            precision=precision,  # Region precision
+            pdb_id=pdb_id,  # PDB ID
+            chain_id=chain_id,  # Chain ID
+            num_nodes=n_nodes,  # Number of nodes
+            embeddings2=node_embeddings2,  # other embeddings [n_nodes, embed_dim] - region-specific
+        )
+        
+        return data
+        
+    except Exception as e:
+        if verbose:
+            print(f"Error creating graph data for {pdb_id}_{chain_id} center {center_idx}: {str(e)}")
+        return None
+    
+def compute_edge_features(coords: np.ndarray, edge_index: torch.Tensor, num_rbf: int = 16, num_posenc: int = 16) -> torch.Tensor:
+    """
+    Compute edge features including RBF and positional encoding.
+        
+    Args:
+        coords: Node coordinates [n_nodes, 3]
+        edge_index: Edge connectivity [2, n_edges]
+        num_rbf: Number of RBF features
+        num_posenc: Number of positional encoding features
+        
+    Returns:
+        Edge features [n_edges, edge_dim]
+        """
+    # Convert to torch tensors
+    coords_tensor = torch.tensor(coords, dtype=torch.float32)
+        
+    # Compute edge vectors and distances
+    edge_vectors = coords_tensor[edge_index[0]] - coords_tensor[edge_index[1]]  # [n_edges, 3]
+    edge_distances = torch.norm(edge_vectors, dim=-1)  # [n_edges]
+        
+    # RBF features
+    edge_rbf = rbf(edge_distances, D_count=num_rbf)  # [n_edges, num_rbf]
+        
+    # Positional encoding
+    edge_posenc = get_posenc(edge_index, num_posenc=num_posenc)  # [n_edges, num_posenc]
+        
+    # Concatenate edge features
+    edge_features = torch.cat([edge_rbf, edge_posenc], dim=-1)  # [n_edges, num_rbf + num_posenc]
+        
+    return edge_features
+
+
+def get_edges(n_nodes):
+    """Generate fully connected edge indices (no self-loops)"""
+    rows, cols = [], []
+    for i in range(n_nodes):
+        for j in range(n_nodes):
+            if i != j:
+                rows.append(i)
+                cols.append(j)
+    return [rows, cols]
+
+
+def get_posenc(edge_index, num_posenc=16):
+    """
+    Generate positional encoding for edges.
+    From https://github.com/jingraham/neurips19-graph-protein-design
+    """
+    d = edge_index[0] - edge_index[1]
+
+    frequency = torch.exp(
+        torch.arange(0, num_posenc, 2, dtype=torch.float32, device=d.device)
+        * -(np.log(10000.0) / num_posenc)
+    )
+
+    angles = d.unsqueeze(-1) * frequency
+    E = torch.cat((torch.cos(angles), torch.sin(angles)), -1)
+    return E
+
+
+def rbf(D, D_min=0., D_max=20., D_count=16):
+    """
+    Radial Basis Function (RBF) encoding for distances.
+    From https://github.com/jingraham/neurips19-graph-protein-design
+
+    Returns an RBF embedding of `torch.Tensor` `D` along a new axis=-1.
+    That is, if `D` has shape [...dims], then the returned tensor will have
+    shape [...dims, D_count].
+    """
+    D_mu = torch.linspace(D_min, D_max, D_count, device=D.device)
+    D_mu = D_mu.view([1, -1])
+    D_sigma = (D_max - D_min) / D_count
+    D_expand = torch.unsqueeze(D, -1)
+
+    RBF = torch.exp(-((D_expand - D_mu) / D_sigma) ** 2)
+    return RBF 

src/bce/model/EGNN.py

@@ -0,0 +1,315 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch_scatter import scatter
+from torch_geometric.nn import InstanceNorm
+
+class EGNNLayer(nn.Module):
+    """
+    EGNN layer with optional feed forward network and batch normalization.
+    
+    Args:
+        input_nf: Number of input node features
+        output_nf: Number of output node features
+        hidden_nf: Number of hidden features
+        edges_in_d: Number of input edge features
+        act_fn: Activation function
+        residual: Whether to use residual connections
+        attention: Whether to use attention mechanism for edge features
+        normalize: Whether to normalize coordinates
+        coords_agg: Aggregation method for coordinates (mean, sum, max, min)
+        tanh: Whether to use tanh activation for coordinate updates
+        dropout: Dropout rate
+        ffn: Whether to use feed forward network
+        batch_norm: Whether to use batch normalization
+    """
+    def __init__(self, input_nf, output_nf, hidden_nf,
+                 edges_in_d=0, act_fn=nn.SiLU(),
+                 residual=True, attention=False, normalize=False,
+                 coords_agg='mean', tanh=False, dropout=0.0,
+                 ffn=False, batch_norm=True):
+        super().__init__()
+        self.input_nf = input_nf
+        self.output_nf = output_nf
+        self.hidden_nf = hidden_nf
+        self.residual = residual
+        self.attention = attention
+        self.normalize = normalize
+        self.coords_agg = coords_agg
+        self.tanh = tanh
+        self.epsilon = 1e-8
+        self.dropout = dropout
+        self.ffn = ffn
+        self.batch_norm = batch_norm
+
+        # Edge MLP
+        in_edge = input_nf*2 + 1 + edges_in_d
+        self.edge_mlp = nn.Sequential(
+            nn.Linear(in_edge, hidden_nf),
+            act_fn, nn.Dropout(dropout),
+            nn.Linear(hidden_nf, hidden_nf),
+            act_fn, nn.Dropout(dropout),
+        )
+        if attention:
+            self.att_mlp = nn.Sequential(nn.Linear(hidden_nf,1), nn.Sigmoid())
+
+        # Coord MLP
+        layer = nn.Linear(hidden_nf,1, bias=False)
+        nn.init.xavier_uniform_(layer.weight, gain=0.001)
+        coord_blocks = [nn.Linear(hidden_nf, hidden_nf), act_fn,
+                        nn.Dropout(dropout), layer]
+        if tanh: coord_blocks.append(nn.Tanh())
+        self.coord_mlp = nn.Sequential(*coord_blocks)
+
+        # Node MLP
+        self.node_mlp = nn.Sequential(
+            nn.Linear(hidden_nf + input_nf, hidden_nf),
+            act_fn, nn.Dropout(dropout),
+            nn.Linear(hidden_nf, output_nf),
+        )
+
+        # per-graph normalization
+        if batch_norm:
+            self.norm_node = InstanceNorm(output_nf, affine=True)
+            self.norm_coord = InstanceNorm(3, affine=True)
+
+        # FFN
+        if ffn:
+            self.ff1 = nn.Linear(output_nf, output_nf*2)
+            self.ff2 = nn.Linear(output_nf*2, output_nf)
+            self.act_ff = act_fn
+            self.drop_ff = nn.Dropout(dropout)
+            if batch_norm:
+                self.norm_ff1 = InstanceNorm(output_nf, affine=True)
+                self.norm_ff2 = InstanceNorm(output_nf, affine=True)
+
+    def coord2radial(self, edge_index, coord):
+        row, col = edge_index
+        diff = coord[row] - coord[col]
+        dist2 = (diff**2).sum(dim=-1, keepdim=True)
+        
+        # Clamp distance to prevent extreme values
+        dist2 = torch.clamp(dist2, min=self.epsilon, max=100.0)
+        
+        if self.normalize:
+            norm = (dist2.sqrt().detach() + self.epsilon)
+            diff = diff / norm
+            # Check for NaN/Inf in normalized diff
+            diff = torch.where(torch.isfinite(diff), diff, torch.zeros_like(diff))
+        return dist2, diff
+
+    def _ff_block(self, x):
+        """Feed Forward block.
+        """
+        x = self.drop_ff(self.act_ff(self.ff1(x)))
+        return self.ff2(x)
+    
+    def forward(self, h, coord, edge_index, batch, edge_attr=None, node_attr=None):
+        row, col = edge_index
+        radial, coord_diff = self.coord2radial(edge_index, coord)
+
+        # -- edge features --
+        e_in = [h[row], h[col], radial]
+        if edge_attr is not None: e_in.append(edge_attr)
+        e = torch.cat(e_in, dim=-1)
+        e = self.edge_mlp(e)
+        if self.attention:
+            att = self.att_mlp(e)
+            e = e * att
+
+        # -- coordinate update --
+        coord_update = self.coord_mlp(e)  # [E,1]
+        # Clamp coordinate updates to prevent explosion
+        coord_update = torch.clamp(coord_update, -1.0, 1.0)
+        trans = coord_diff * coord_update  # [E,3]
+        
+        # Check for NaN/Inf in coordinate updates
+        trans = torch.where(torch.isfinite(trans), trans, torch.zeros_like(trans))
+        
+        agg_coord = scatter(trans, row, dim=0,
+                            dim_size=coord.size(0),
+                            reduce=self.coords_agg)
+        coord = coord + agg_coord
+        
+        # Check for NaN/Inf in final coordinates
+        coord = torch.where(torch.isfinite(coord), coord, torch.zeros_like(coord))
+        
+        if self.batch_norm:
+            coord = self.norm_coord(coord, batch)
+
+        # -- node update --
+        agg_node = scatter(e, row, dim=0,
+                           dim_size=h.size(0), reduce='sum')
+        x_in = torch.cat([h, agg_node], dim=-1)
+        if node_attr is not None:
+            x_in = torch.cat([x_in, node_attr], dim=-1)
+        h_new = self.node_mlp(x_in)
+        if self.batch_norm:
+            h_new = self.norm_node(h_new, batch)
+        if self.residual and h_new.shape[-1] == h.shape[-1]:
+            h_new = h + h_new
+
+        # -- optional FFN --
+        if self.ffn:
+            if self.batch_norm:
+                h_new = self.norm_ff1(h_new, batch)
+            h_new = h_new + self._ff_block(h_new)
+            if self.batch_norm:
+                h_new = self.norm_ff2(h_new, batch)
+
+        return h_new, coord, e
+
+class EGNNLayer2(nn.Module):
+    """
+    EGNN layer with optional feed forward network and batch normalization.
+    
+    Args:
+        input_nf: Number of input node features
+        output_nf: Number of output node features
+        hidden_nf: Number of hidden features
+        edges_in_d: Number of input edge features
+        act_fn: Activation function
+        residual: Whether to use residual connections
+        attention: Whether to use attention mechanism for edge features
+        normalize: Whether to normalize coordinates
+        coords_agg: Aggregation method for coordinates (mean, sum, max, min)
+        tanh: Whether to use tanh activation for coordinate updates
+        dropout: Dropout rate
+        ffn: Whether to use feed forward network
+        batch_norm: Whether to use batch normalization
+    """
+    def __init__(self, input_nf, output_nf, hidden_nf,
+                 edges_in_d=0, act_fn=nn.SiLU(),
+                 residual=True, attention=False, normalize=False,
+                 coords_agg='mean', tanh=False, dropout=0.0,
+                 ffn=False, batch_norm=True):
+        super().__init__()
+        self.input_nf = input_nf
+        self.output_nf = output_nf
+        self.hidden_nf = hidden_nf
+        self.residual = residual
+        self.attention = attention
+        self.normalize = normalize
+        self.coords_agg = coords_agg
+        self.tanh = tanh
+        self.epsilon = 1e-8
+        self.dropout = dropout
+        self.ffn = ffn
+        self.batch_norm = batch_norm
+
+        # Edge MLP
+        in_edge = input_nf*2 + 1 + edges_in_d
+        self.edge_mlp = nn.Sequential(
+            nn.Linear(in_edge, hidden_nf),
+            act_fn, nn.Dropout(dropout),
+            nn.Linear(hidden_nf, hidden_nf),
+            act_fn, nn.Dropout(dropout),
+        )
+        if attention:
+            self.att_mlp = nn.Sequential(nn.Linear(hidden_nf,1), nn.Sigmoid())
+
+        # Coord MLP
+        layer = nn.Linear(hidden_nf,1, bias=False)
+        nn.init.xavier_uniform_(layer.weight, gain=0.001)
+        coord_blocks = [nn.Linear(hidden_nf, hidden_nf), act_fn,
+                        nn.Dropout(dropout), layer]
+        if tanh: coord_blocks.append(nn.Tanh())
+        self.coord_mlp = nn.Sequential(*coord_blocks)
+
+        # Node MLP
+        self.node_mlp = nn.Sequential(
+            nn.Linear(hidden_nf + input_nf, hidden_nf),
+            act_fn, nn.Dropout(dropout),
+            nn.Linear(hidden_nf, output_nf),
+        )
+
+        # per-graph normalization
+        if batch_norm:
+            self.norm_node = InstanceNorm(output_nf, affine=True)
+            self.norm_coord = InstanceNorm(3, affine=True)
+
+        # FFN
+        if ffn:
+            self.ff1 = nn.Linear(output_nf, output_nf*2)
+            self.ff2 = nn.Linear(output_nf*2, output_nf)
+            self.act_ff = act_fn
+            self.drop_ff = nn.Dropout(dropout)
+            if batch_norm:
+                self.norm_ff1 = InstanceNorm(output_nf, affine=True)
+                self.norm_ff2 = InstanceNorm(output_nf, affine=True)
+
+    def coord2radial(self, edge_index, coord):
+        row, col = edge_index
+        diff = coord[row] - coord[col]
+        dist2 = (diff**2).sum(dim=-1, keepdim=True)
+        
+        # Clamp distance to prevent extreme values
+        dist2 = torch.clamp(dist2, min=self.epsilon, max=100.0)
+        
+        if self.normalize:
+            norm = (dist2.sqrt().detach() + self.epsilon)
+            diff = diff / norm
+            # Check for NaN/Inf in normalized diff
+            diff = torch.where(torch.isfinite(diff), diff, torch.zeros_like(diff))
+        return dist2, diff
+
+    def _ff_block(self, x):
+        """Feed Forward block.
+        """
+        x = self.drop_ff(self.act_ff(self.ff1(x)))
+        return self.ff2(x)
+    
+    def forward(self, h, coord, edge_index, batch, edge_attr=None, node_attr=None):
+        row, col = edge_index
+        radial, coord_diff = self.coord2radial(edge_index, coord)
+
+        # -- edge features --
+        e_in = [h[row], h[col], radial]
+        if edge_attr is not None: e_in.append(edge_attr)
+        e = torch.cat(e_in, dim=-1)
+        e = self.edge_mlp(e)
+        if self.attention:
+            att = self.att_mlp(e)
+            e = e * att
+
+        # -- coordinate update --
+        coord_update = self.coord_mlp(e)  # [E,1]
+        # Clamp coordinate updates to prevent explosion
+        coord_update = torch.clamp(coord_update, -1.0, 1.0)
+        trans = coord_diff * coord_update  # [E,3]
+        
+        # Check for NaN/Inf in coordinate updates
+        trans = torch.where(torch.isfinite(trans), trans, torch.zeros_like(trans))
+        
+        agg_coord = scatter(trans, row, dim=0,
+                            dim_size=coord.size(0),
+                            reduce=self.coords_agg)
+        coord = coord + agg_coord
+        
+        # Check for NaN/Inf in final coordinates
+        coord = torch.where(torch.isfinite(coord), coord, torch.zeros_like(coord))
+        
+        if self.batch_norm:
+            coord = self.norm_coord(coord, batch)
+
+        # -- node update --
+        agg_node = scatter(e, row, dim=0,
+                           dim_size=h.size(0), reduce='sum')
+        x_in = torch.cat([h, agg_node], dim=-1)
+        if node_attr is not None:
+            x_in = torch.cat([x_in, node_attr], dim=-1)
+        h_new = self.node_mlp(x_in)
+        if self.batch_norm:
+            h_new = self.norm_node(h_new, batch)
+        if self.residual and h_new.shape[-1] == h.shape[-1]:
+            h_new = h + h_new
+
+        # -- optional FFN --
+        if self.ffn:
+            if self.batch_norm:
+                h_new = self.norm_ff1(h_new, batch)
+            h_new = h_new + self._ff_block(h_new)
+            if self.batch_norm:
+                h_new = self.norm_ff2(h_new, batch)
+
+        return h_new, coord, e

src/bce/model/ReCEP.py

@@ -0,0 +1,481 @@
+from pathlib import Path
+from prettytable import PrettyTable
+
+import torch
+import torch.nn as nn
+from torch_scatter import scatter_softmax, scatter_sum
+from torch_geometric.data import Data, Batch
+
+from .dihedral import DihedralFeatures
+from .EGNN import EGNNLayer
+from .pooling import AttentionPooling, AddPooling
+from .activation import get_activation
+# from .baseline import EP
+
+class ReCEP(nn.Module):
+    """
+    Refined Graph Epitope Predictor with optional EGNN layer skipping for ablation.
+    """
+    def __init__(
+        self,
+        in_dim: int = 2560,
+        rsa: bool = True,
+        dihedral: bool = True,
+        node_dims: list = [512, 256, 256],
+        edge_dim: int = 32,
+        dropout: float = 0.3,
+        activation: str = "gelu",
+        residual: bool = True,
+        attention: bool = True,
+        normalize: bool = True,
+        coords_agg: str = 'mean',
+        ffn: bool = True,
+        batch_norm: bool = True,
+        concat: bool = False,
+        addition: bool = False,
+        # Global predictor
+        pooling: str = 'attention',
+        # Node classifier
+        fusion_type: str = 'concat',
+        node_gate: bool = False,
+        node_norm: bool = False,
+        node_layers: int = 2,
+        out_dropout: float = 0.2,
+        use_egnn: bool = True,  # NEW: toggle for EGNN layer usage
+        encoder: str = 'esmc',
+    ):
+        super().__init__()
+        self.use_egnn = use_egnn
+        self.in_dim = in_dim
+        self.rsa = rsa
+        self.dihedral = dihedral
+        self.original_node_dims = node_dims.copy()
+        self.edge_dim = edge_dim
+        self.dropout = dropout
+        self.activation = activation
+        self.residual = residual
+        self.attention = attention
+        self.normalize = normalize
+        self.ffn = ffn
+        self.batch_norm = batch_norm
+        self.coords_agg = coords_agg
+        self.concat = concat
+        self.addition = addition
+        self.fusion_type = fusion_type
+        self.node_gate = node_gate
+        self.node_norm = node_norm
+        self.node_layers = node_layers
+        self.out_dropout = out_dropout
+        self.pooling = pooling
+
+        self.base_node_dim = node_dims[0]
+        self.node_dims = node_dims.copy()
+        self.node_dims[0] += 1 if rsa else 0
+        
+        self.node_dims[-1] = self.node_dims[0] if addition else self.node_dims[-1]
+        
+        # Modify input dimension based on encoder
+        self.encoder = encoder
+        if encoder == 'esmc':
+            self.in_dim = 2560
+        elif encoder == 'esm2':
+            self.in_dim = 1280
+        else:
+            self.in_dim = in_dim
+        
+        # Calculate actual final node dimension based on whether EGNN is used
+        if self.use_egnn:
+            self.final_node_dim = self.node_dims[-1]
+        else:
+            self.final_node_dim = self.node_dims[0]
+            self.concat = False
+            self.addition = False
+
+        self.proj_layer = nn.Sequential(
+            nn.Linear(self.in_dim, self.base_node_dim),
+            get_activation(activation),
+            nn.Dropout(dropout),
+        )
+
+        if dihedral:
+            try:
+                self.dihedral_features = DihedralFeatures(self.base_node_dim)
+            except:
+                print("Warning: DihedralFeatures not found, skipping dihedral features")
+                self.dihedral = False
+
+        self.egnn_layers = nn.ModuleList()
+        if self.use_egnn:
+            for i in range(len(self.node_dims) - 1):
+                self.egnn_layers.append(
+                    EGNNLayer(
+                        input_nf=self.node_dims[i],
+                        output_nf=self.node_dims[i+1],
+                        hidden_nf=self.node_dims[i+1],
+                        edges_in_d=edge_dim,
+                        act_fn=get_activation(activation),
+                        residual=residual,
+                        attention=attention,
+                        normalize=normalize,
+                        coords_agg=coords_agg,
+                        dropout=dropout,
+                        ffn=ffn,
+                        batch_norm=batch_norm
+                    )
+                )
+
+        if concat and self.use_egnn:
+            self.final_node_dim += self.node_dims[0]
+            
+        if addition and self.use_egnn:
+            assert self.node_dims[0] == self.node_dims[-1], "Node dimension mismatch for addition"
+            self.final_node_dim = self.node_dims[0]
+            
+        # Calculate node classifier input dimension based on fusion type
+        if fusion_type == 'concat':
+            self.node_classifier_input_dim = self.final_node_dim * 2
+        elif fusion_type == 'add':
+            self.node_classifier_input_dim = self.final_node_dim
+        else:
+            raise ValueError(f"Unsupported fusion type: {fusion_type}")
+        
+        # Calculate node gate input dimension
+        if node_gate:
+            if fusion_type == 'concat':
+                self.node_gate_input_dim = self.final_node_dim * 2
+            elif fusion_type == 'add':
+                self.node_gate_input_dim = self.final_node_dim
+            else:
+                raise ValueError(f"Unsupported fusion type: {fusion_type}")
+
+        if pooling == 'attention':
+            self.graph_pool = AttentionPooling(
+                input_dim=self.final_node_dim,
+                dropout=dropout,
+                activation=activation
+            )
+        elif pooling == 'add':
+            self.graph_pool = AddPooling(
+                input_dim=self.final_node_dim,
+                dropout=dropout
+            )
+        else:
+            raise ValueError(f"Unsupported pooling method: {pooling}")
+
+        self.global_predictor = nn.Sequential(
+            nn.Linear(self.final_node_dim, self.final_node_dim // 2),
+            get_activation(activation),
+            nn.Dropout(out_dropout),
+            nn.Linear(self.final_node_dim // 2, 1)
+        )
+
+        if node_gate:
+            self.node_gate = nn.Sequential(
+                nn.Linear(self.node_gate_input_dim, self.final_node_dim),
+                get_activation(activation),
+                nn.LayerNorm(self.final_node_dim),
+                nn.Linear(self.final_node_dim, self.final_node_dim),
+                nn.Sigmoid()
+            )
+
+        self.node_classifier = self._build_node_classifier()
+
+        self._param_printed = False
+        self.apply(self._init_weights)
+
+    def _build_node_classifier(self):
+        layers = []
+        input_dim = self.node_classifier_input_dim
+        current_dim = input_dim
+        for i in range(self.node_layers):
+            output_dim = 1 if i == self.node_layers - 1 else max(current_dim // 2, 32)
+            layers.append(nn.Linear(current_dim, output_dim))
+            if self.node_norm and i < self.node_layers - 1:
+                layers.append(nn.LayerNorm(output_dim))
+            if i < self.node_layers - 1:
+                layers.append(get_activation(self.activation))
+                layers.append(nn.Dropout(self.out_dropout))
+            current_dim = output_dim
+        return nn.Sequential(*layers)
+
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            nn.init.xavier_normal_(module.weight)
+            if module.bias is not None:
+                nn.init.constant_(module.bias, 0.0 if module.out_features == 1 else 0.01)
+        elif isinstance(module, nn.LayerNorm):
+            nn.init.ones_(module.weight)
+            nn.init.zeros_(module.bias)
+
+    def forward(self, data: Batch) -> dict:
+        if self.training and not self._param_printed:
+            print(f"ReCEP total params: {sum(p.numel() for p in self.parameters()):,}")
+            self._param_printed = True
+
+        x = data.x
+    
+        coords = data.pos
+        batch = data.batch
+        e_attr = data.edge_attr
+        coords_C = coords[:, 1].clone()
+
+        x = self.proj_layer(x)
+        if self.dihedral and coords is not None:
+            x = x + self.dihedral_features(coords)
+        if self.rsa and data.rsa is not None:
+            rsa = data.rsa.unsqueeze(-1)
+            x = torch.cat([x, rsa], dim=-1)
+
+        h = x
+        assert h.shape[1] == self.node_dims[0], f"[ReCEP] Node feature dim mismatch: got {h.shape[1]}, expected {self.node_dims[0]}"
+
+        if self.use_egnn:
+            for layer in self.egnn_layers:
+                h, coords_C, _ = layer(h, coords_C, data.edge_index, batch, edge_attr=e_attr)
+
+        if self.concat and self.use_egnn:
+            h = torch.cat([x, h], dim=-1)
+        elif self.addition and self.use_egnn:
+            h = h + x
+
+        graph_feats = self.graph_pool(h, batch)
+        global_pred = self.global_predictor(graph_feats).squeeze(-1)
+
+        context = graph_feats[batch]
+        if self.node_gate and hasattr(self, 'node_gate'):
+            if self.fusion_type == 'concat':
+                gate_input = torch.cat([h, context], dim=-1)
+            elif self.fusion_type == 'add':
+                gate_input = h + context
+            else:
+                raise ValueError(f"Unsupported fusion type: {self.fusion_type}")
+            gate = self.node_gate(gate_input)
+            gated_h = h + gate * h
+        else:
+            gated_h = h
+
+        if self.fusion_type == 'concat':
+            cat = torch.cat([gated_h, context], dim=-1)
+        elif self.fusion_type == 'add':
+            # Ensure dimensions match for addition
+            assert gated_h.shape[-1] == context.shape[-1], f"[ReCEP] Dimension mismatch for add fusion: gated_h {gated_h.shape[-1]} vs context {context.shape[-1]}"
+            cat = gated_h + context
+        else:
+            raise ValueError(f"Unsupported fusion type: {self.fusion_type}")
+        
+        # Verify input dimension matches node classifier expectation
+        expected_dim = self.node_classifier_input_dim
+        actual_dim = cat.shape[-1]
+        assert actual_dim == expected_dim, f"[ReCEP] Node classifier input dim mismatch: got {actual_dim}, expected {expected_dim}"
+        
+        node_preds = self.node_classifier(cat).squeeze(-1)
+
+        return {"global_pred": global_pred, "node_preds": node_preds}
+    
+    def print_param_count(self):
+        """Print a summary table of parameter counts"""
+        table = PrettyTable()
+        table.field_names = ["Layer Name", "Type", "Parameters", "Trainable"]
+        total_params = 0
+        trainable_params = 0
+
+        for name, module in self.named_modules():
+            if not list(module.children()):  # Only leaf nodes
+                params = sum(p.numel() for p in module.parameters())
+                is_trainable = any(p.requires_grad for p in module.parameters())
+                
+                if params > 0:
+                    total_params += params
+                    trainable_params += params if is_trainable else 0
+                    
+                    table.add_row([
+                        name,
+                        module.__class__.__name__,
+                        f"{params:,}",
+                        "✓" if is_trainable else "✗"
+                    ])
+
+        table.add_row(["", "", "", ""], divider=True)
+        table.add_row([
+            "TOTAL", 
+            "", 
+            f"{total_params:,}",
+            f"Trainable: {trainable_params:,}"
+        ])
+
+        print("\nReCEP Model Parameter Summary:")
+        print(table)
+        print(f"Parameter Density: {trainable_params/total_params:.1%}\n")
+    
+    def save(self, path, threshold: float = 0.5):
+        """Save model with configuration"""
+        path = Path(path)
+        try:
+            path.parent.mkdir(parents=True, exist_ok=True)
+            save_path = path.with_suffix('.bin')
+            
+            config = self.get_config()
+            # config = {
+            #     'in_dim': self.in_dim,
+            #     'rsa': self.rsa,
+            #     'dihedral': self.dihedral,
+            #     'node_dims': self.original_node_dims,  # Use original node_dims
+            #     'edge_dim': self.edge_dim,
+            #     'dropout': self.dropout,
+            #     'activation': self.activation,
+            #     'residual': self.residual,
+            #     'attention': self.attention,
+            #     'normalize': self.normalize,
+            #     'coords_agg': self.coords_agg,
+            #     'ffn': self.ffn,
+            #     'batch_norm': self.batch_norm,
+            #     'concat': self.concat,
+            #     'node_norm': self.node_norm,
+            #     'node_layers': self.node_layers,
+            #     'node_gate': self.node_gate,
+            #     'out_dropout': self.out_dropout
+            # }
+            
+            torch.save({
+                'model_state': self.state_dict(),
+                'config': config,
+                'model_class': self.__class__.__name__,
+                'version': '1.0',
+                'threshold': threshold
+            }, save_path)
+            print(f"ReCEP model saved to {save_path}")
+        except Exception as e:
+            print(f"Save failed: {str(e)}")
+            raise
+    
+    @classmethod
+    def load(cls, path, device='cpu', strict=True, verbose=True):
+        """Load model with configuration"""
+        path = Path(path)
+        if not path.exists():
+            raise FileNotFoundError(f"Model file {path} not found")
+
+        try:
+            if isinstance(device, str):
+                device = torch.device(device)
+            elif isinstance(device, int):
+                if device >= 0 and torch.cuda.is_available():
+                    device = torch.device(f'cuda:{device}')
+                else:
+                    device = torch.device('cpu')
+            elif not isinstance(device, torch.device):
+                raise ValueError(f"Unsupported device type: {type(device)}")
+            
+            checkpoint = torch.load(
+                path,
+                map_location=device,
+                weights_only=False
+            )
+        except RuntimeError:
+            print("Warning: Using unsafe load due to weights_only restriction")
+            checkpoint = torch.load(path, map_location=device)
+
+        # Version compatibility check
+        if 'version' not in checkpoint:
+            print("Warning: Loading legacy model without version info")
+        
+        # Rebuild configuration
+        config = checkpoint.get('config', {})
+        model = cls(**config)
+        
+        # Load state dict
+        model_state = checkpoint['model_state']
+        current_state = model.state_dict()
+        
+        # Auto-match parameters
+        matched_state = {}
+        for name, param in model_state.items():
+            if name in current_state:
+                if param.shape == current_state[name].shape:
+                    matched_state[name] = param
+                else:
+                    print(f"Size mismatch: {name} (load {param.shape} vs model {current_state[name].shape})")
+            else:
+                print(f"Parameter not found: {name}")
+        
+        current_state.update(matched_state)
+        model.load_state_dict(current_state, strict=strict)
+        
+        if verbose:
+            print(f"Successfully loaded {len(matched_state)}/{len(model_state)} parameters")
+        
+        return model.to(device), checkpoint.get('threshold', 0.5)
+    
+    def get_config(self):
+        """Get model configuration"""
+        return {
+            'in_dim': self.in_dim,
+            'rsa': self.rsa,
+            'dihedral': self.dihedral,
+            'node_dims': self.original_node_dims,
+            'edge_dim': self.edge_dim,
+            'dropout': self.dropout,
+            'activation': self.activation,
+            'residual': self.residual,
+            'attention': self.attention,
+            'normalize': self.normalize,
+            'coords_agg': self.coords_agg,
+            'ffn': self.ffn,
+            'batch_norm': self.batch_norm,
+            'concat': self.concat,
+            'addition': self.addition,
+            'pooling': self.pooling,
+            'fusion_type': self.fusion_type,
+            'node_gate': self.node_gate,
+            'node_norm': self.node_norm,
+            'node_layers': self.node_layers,
+            'out_dropout': self.out_dropout,
+            'use_egnn': self.use_egnn,
+            'encoder': self.encoder
+        }
+        
+        
+model_registry = {
+    "ReCEP": ReCEP,
+}
+
+def get_model(configs):
+    """
+    Flexible model loader. Accepts either an argparse.Namespace or a dict.
+    Returns an instance of the selected model.
+    """
+    # Support both argparse.Namespace and dict
+    if hasattr(configs, '__dict__'):
+        args = vars(configs)
+    else:
+        args = configs
+    
+    # Default to ReCEP if no model specified
+    model_name = args.get('model', 'ReCEP')
+    
+    if model_name not in model_registry:
+        valid_models = list(model_registry.keys())
+        raise ValueError(f"Invalid model type: {model_name}. Must be one of: {valid_models}")
+    
+    model_class = model_registry[model_name]
+    
+    # Use inspect to get the model's __init__ parameters
+    import inspect
+    init_signature = inspect.signature(model_class.__init__)
+    parameters = init_signature.parameters
+    
+    # Build model configuration from args
+    model_config = {}
+    for param_name, param in parameters.items():
+        if param_name == 'self':
+            continue
+        if param_name in args:
+            model_config[param_name] = args[param_name]
+        elif param.default is not param.empty:
+            model_config[param_name] = param.default
+        else:
+            print(f"[WARNING] Required parameter '{param_name}' not found in args and has no default value")
+    
+    # print(f"[INFO] Creating {model_name} model with config: {list(model_config.keys())}")
+    model = model_class(**model_config)
+    return model

src/bce/model/__init__.py

@@ -0,0 +1,11 @@
+"""
+BCE Model Module
+
+This module contains various neural network models for B-cell epitope prediction.
+"""
+
+# from .ReCEP import ReCEP
+# from .EGNN import EGNNLayer
+# from .dihedral import DihedralFeatures
+
+# __all__ = ['ReCEP', 'EGNNLayer', 'DihedralFeatures']