Update mcp/orchestrator.py

mcp/orchestrator.py

@@ -1,127 +1,80 @@
-"""
-MedGenesis – dual-LLM orchestrator
-----------------------------------
-• Accepts llm = "openai" | "gemini"   (falls back to OpenAI)
-• Returns one unified dict the UI can rely on.
-"""
-from __future__ import annotations
-import asyncio, itertools, logging
-from typing import Dict, Any, List, Tuple
-
-from mcp.arxiv            import fetch_arxiv
-from mcp.pubmed           import fetch_pubmed
-from mcp.ncbi             import search_gene, get_mesh_definition
-from mcp.mygene           import fetch_gene_info
-from mcp.ensembl          import fetch_ensembl
-from mcp.opentargets      import fetch_ot
-from mcp.umls             import lookup_umls
-from mcp.openfda          import fetch_drug_safety
-from mcp.disgenet         import disease_to_genes
-from mcp.clinicaltrials   import search_trials
-from mcp.cbio             import fetch_cbio
-from mcp.openai_utils     import ai_summarize, ai_qa
-from mcp.gemini           import gemini_summarize, gemini_qa
-
-log = logging.getLogger(__name__)
-_DEF = "openai"                                # default engine
-
-
-# ─────────────────────────────────── helpers ───────────────────────────────────
-def _llm_router(engine: str = _DEF) -> Tuple:
-    if engine.lower() == "gemini":
-        return gemini_summarize, gemini_qa, "gemini"
-    return ai_summarize, ai_qa, "openai"
-
-async def _gather_safely(*aws, as_list: bool = True):
-    """await gather() that converts Exception → RuntimeError placeholder"""
-    out = await asyncio.gather(*aws, return_exceptions=True)
-    if as_list:
-        # filter exceptions – keep structure but drop failures
-        return [x for x in out if not isinstance(x, Exception)]
-    return out
-
-async def _gene_enrichment(keys: List[str]) -> Dict[str, Any]:
-    jobs = []
-    for k in keys:
-        jobs += [
-            search_gene(k),                    # basic gene info
-            get_mesh_definition(k),            # MeSH definitions
-            fetch_gene_info(k),                # MyGene
-            fetch_ensembl(k),                  # Ensembl x-refs
-            fetch_ot(k),                       # Open Targets associations
-        ]
-    res = await _gather_safely(*jobs, as_list=False)
-
-    # slice & compress five-way fan-out
-    combo = lambda idx: [r for i, r in enumerate(res) if i % 5 == idx and r]
-    return {
-        "ncbi"     : combo(0),
-        "mesh"     : combo(1),
-        "mygene"   : combo(2),
-        "ensembl"  : combo(3),
-        "ot_assoc" : combo(4),
-    }
-
-
-# ───────────────────────────────── orchestrator ────────────────────────────────
-async def orchestrate_search(query: str, *, llm: str = _DEF) -> Dict[str, Any]:
-    """Main entry – returns dict for the Streamlit UI"""
-    # 1  Literature – run in parallel
-    arxiv_task  = asyncio.create_task(fetch_arxiv(query))
-    pubmed_task = asyncio.create_task(fetch_pubmed(query))
-    papers_raw  = await _gather_safely(arxiv_task, pubmed_task)
-    papers      = list(itertools.chain.from_iterable(papers_raw))[:30]   # keep ≤30
-
-    # 2  Keyword extraction (very light – only from abstracts)
-    kws = {w for p in papers for w in (p["summary"][:500].split()) if w.isalpha()}
-    kws = list(kws)[:10]                           # coarse, fast -> 10 seeds
-
-    # 3  Bio-enrichment fan-out
-    umls_f       = [_safe_task(lookup_umls, k) for k in kws]
-    fda_f        = [_safe_task(fetch_drug_safety, k) for k in kws]
-    gene_bundle  = asyncio.create_task(_gene_enrichment(kws))
-    trials_task  = asyncio.create_task(search_trials(query, max_studies=20))
-    cbio_task    = asyncio.create_task(fetch_cbio(kws[0] if kws else ""))
-
-    umls, fda, gene_dat, trials, variants = await asyncio.gather(
-        _gather_safely(*umls_f),
-        _gather_safely(*fda_f),
-        gene_bundle,
-        trials_task,
-        cbio_task,
+# mcp/orchestrator.py
+
+import asyncio
+from mcp.pubmed import fetch_pubmed
+from mcp.arxiv import fetch_arxiv
+from mcp.umls import extract_umls_concepts
+from mcp.openfda import fetch_drug_safety
+from mcp.ncbi import search_gene, get_mesh_definition
+from mcp.mygene import fetch_gene_info
+from mcp.ensembl import fetch_ensembl
+from mcp.opentargets import fetch_ot
+from mcp.clinicaltrials import search_trials
+from mcp.cbio import fetch_cbio
+from mcp.gemini import gemini_summarize, gemini_qa
+from mcp.openai_utils import ai_summarize, ai_qa
+from mcp.disgenet import disease_to_genes
+
+async def orchestrate_search(query, llm="openai"):
+    # --- Literature: PubMed + arXiv
+    pubmed_task = asyncio.create_task(fetch_pubmed(query, max_results=7))
+    arxiv_task = asyncio.create_task(fetch_arxiv(query, max_results=7))
+    # --- UMLS, OpenFDA, Gene, Mesh
+    umls_task = asyncio.create_task(extract_umls_concepts(query))
+    fda_task = asyncio.create_task(fetch_drug_safety(query))
+    gene_ncbi_task = asyncio.create_task(search_gene(query))
+    mygene_task = asyncio.create_task(fetch_gene_info(query))
+    ensembl_task = asyncio.create_task(fetch_ensembl(query))
+    ot_task = asyncio.create_task(fetch_ot(query))
+    mesh_task = asyncio.create_task(get_mesh_definition(query))
+    # --- Trials, cBio, DisGeNET
+    trials_task = asyncio.create_task(search_trials(query, max_studies=10))
+    cbio_task = asyncio.create_task(fetch_cbio(query))
+    disgenet_task = asyncio.create_task(disease_to_genes(query))
+
+    # Run
+    pubmed, arxiv, umls, fda, ncbi, mygene, ensembl, ot, mesh, trials, cbio, disgenet = await asyncio.gather(
+        pubmed_task, arxiv_task, umls_task, fda_task, gene_ncbi_task,
+        mygene_task, ensembl_task, ot_task, mesh_task, trials_task, cbio_task, disgenet_task
     )
-
-    # 4  LLM summary
-    summarise_fn, _, engine = _llm_router(llm)
-    summary = await summarise_fn(" ".join(p["summary"] for p in papers)[:12000])
+    # Genes: flatten and deduplicate
+    genes = []
+    for g in (ncbi, mygene, ensembl, ot):
+        if isinstance(g, list):
+            genes.extend(g)
+        elif isinstance(g, dict) and g:
+            genes.append(g)
+    genes = [g for i, g in enumerate(genes) if g and genes.index(g) == i]  # dedup
+
+    # --- AI summary (LLM engine select)
+    papers = (pubmed or []) + (arxiv or [])
+    if llm == "gemini":
+        ai_summary = await gemini_summarize(" ".join([p.get("summary", "") for p in papers]))
+        llm_used = "gemini"
+    else:
+        ai_summary = await ai_summarize(" ".join([p.get("summary", "") for p in papers]))
+        llm_used = "openai"
 
     return {
-        "papers"          : papers,
-        "umls"            : umls,
-        "drug_safety"     : fda,
-        "ai_summary"      : summary,
-        "llm_used"        : engine,
-        "genes"           : gene_dat["ncbi"] + gene_dat["ensembl"] + gene_dat["mygene"],
-        "mesh_defs"       : gene_dat["mesh"],
-        "gene_disease"    : gene_dat["ot_assoc"],
-        "clinical_trials" : trials,
-        "variants"        : variants or [],
+        "papers": papers,
+        "genes": genes,
+        "umls": umls or [],
+        "gene_disease": disgenet if isinstance(disgenet, list) else [],
+        "mesh_defs": [mesh] if isinstance(mesh, str) and mesh else [],
+        "drug_safety": fda or [],
+        "clinical_trials": trials or [],
+        "variants": cbio if isinstance(cbio, list) else [],
+        "ai_summary": ai_summary,
+        "llm_used": llm_used
     }
 
-# ─────────────────────────────── follow-up QA ─────────────────────────────────
-async def answer_ai_question(question: str, *, context: str, llm: str = _DEF) -> Dict[str, str]:
-    """Follow-up QA using chosen LLM."""
-    _, qa_fn, _ = _llm_router(llm)
-    return {"answer": await qa_fn(f"Q: {question}\nContext: {context}\nA:")}
-
-
-# ─────────────────────────── internal util  ───────────────────────────────────
-def _safe_task(fn, *args):
-    """Helper to wrap callable → Task returning RuntimeError on exception."""
-    async def _wrapper():
-        try:
-            return await fn(*args)
-        except Exception as exc:
-            log.warning("background task %s failed: %s", fn.__name__, exc)
-            return RuntimeError(str(exc))
-    return asyncio.create_task(_wrapper())
+async def answer_ai_question(question, context="", llm="openai"):
+    # Gemini fallback if OpenAI quota is exceeded
+    try:
+        if llm == "gemini":
+            answer = await gemini_qa(question, context)
+        else:
+            answer = await ai_qa(question, context)
+    except Exception as e:
+        answer = f"LLM unavailable or quota exceeded. ({e})"
+    return {"answer": answer}