Update app.py
Browse files
app.py
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
#!/usr/bin/env python3
|
2 |
-
# MedGenesis AI · CPU-only Streamlit
|
3 |
|
4 |
-
import
|
5 |
-
import asyncio, re
|
6 |
from pathlib import Path
|
7 |
|
8 |
import streamlit as st
|
@@ -11,25 +11,29 @@ import plotly.express as px
|
|
11 |
from fpdf import FPDF
|
12 |
from streamlit_agraph import agraph
|
13 |
|
14 |
-
from mcp.orchestrator
|
15 |
-
from mcp.workspace
|
16 |
from mcp.knowledge_graph import build_agraph
|
17 |
-
from mcp.
|
18 |
-
from mcp.alerts
|
19 |
|
20 |
-
# ── Streamlit telemetry dir fix
|
21 |
-
os.environ["STREAMLIT_DATA_DIR"]
|
22 |
-
os.environ["XDG_STATE_HOME"]
|
23 |
os.environ["STREAMLIT_BROWSER_GATHERUSAGESTATS"] = "false"
|
24 |
pathlib.Path("/tmp/.streamlit").mkdir(parents=True, exist_ok=True)
|
25 |
|
26 |
ROOT = Path(__file__).parent
|
27 |
LOGO = ROOT / "assets" / "logo.png"
|
28 |
|
|
|
|
|
|
|
29 |
def _latin1_safe(txt: str) -> str:
|
30 |
return txt.encode("latin-1", "replace").decode("latin-1")
|
31 |
|
32 |
-
|
|
|
33 |
pdf = FPDF()
|
34 |
pdf.set_auto_page_break(auto=True, margin=15)
|
35 |
pdf.add_page()
|
@@ -45,34 +49,41 @@ def _pdf(papers):
|
|
45 |
pdf.ln(1)
|
46 |
return pdf.output(dest="S").encode("latin-1", "replace")
|
47 |
|
48 |
-
|
|
|
49 |
with st.sidebar:
|
50 |
st.header("🗂️ Workspace")
|
51 |
ws = get_workspace()
|
52 |
if not ws:
|
53 |
st.info("Run a search then press **Save** to populate this list.")
|
54 |
return
|
|
|
|
|
|
|
55 |
for i, item in enumerate(ws, 1):
|
56 |
with st.expander(f"{i}. {item['query']}"):
|
57 |
st.write(item["result"]["ai_summary"])
|
58 |
|
59 |
-
|
|
|
|
|
|
|
|
|
60 |
st.set_page_config("MedGenesis AI", layout="wide")
|
61 |
|
62 |
-
#
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
st.session_state.
|
71 |
-
if "last_llm" not in st.session_state:
|
72 |
-
st.session_state.last_llm = ""
|
73 |
|
74 |
_workspace_sidebar()
|
75 |
|
|
|
76 |
c1, c2 = st.columns([0.15, 0.85])
|
77 |
with c1:
|
78 |
if LOGO.exists():
|
@@ -81,13 +92,14 @@ def render_ui():
|
|
81 |
st.markdown("## 🧬 **MedGenesis AI**")
|
82 |
st.caption("Multi-source biomedical assistant · OpenAI / Gemini")
|
83 |
|
84 |
-
llm
|
85 |
-
query = st.text_input("Enter biomedical question",
|
|
|
86 |
|
87 |
-
#
|
88 |
-
if get_workspace():
|
89 |
try:
|
90 |
-
news = asyncio.run(check_alerts([w["query"] for w in
|
91 |
if news:
|
92 |
with st.sidebar:
|
93 |
st.subheader("🔔 New papers")
|
@@ -96,113 +108,143 @@ def render_ui():
|
|
96 |
except Exception:
|
97 |
pass
|
98 |
|
99 |
-
#
|
100 |
-
if st.button("Run Search 🚀") and query:
|
101 |
with st.spinner("Collecting literature & biomedical data …"):
|
102 |
res = asyncio.run(orchestrate_search(query, llm=llm))
|
103 |
st.success(f"Completed with **{res['llm_used'].title()}**")
|
104 |
-
st.session_state.
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
|
|
|
|
109 |
|
110 |
res = st.session_state.query_result
|
111 |
-
|
112 |
-
if res:
|
113 |
-
tabs = st.tabs(["Results", "Genes", "Trials", "Graph", "Metrics", "Visuals"])
|
114 |
-
|
115 |
-
with tabs[0]:
|
116 |
-
for i, p in enumerate(res["papers"], 1):
|
117 |
-
st.markdown(f"**{i}. [{p['title']}]({p['link']})** *{p['authors']}*")
|
118 |
-
st.write(p["summary"])
|
119 |
-
col1, col2 = st.columns(2)
|
120 |
-
with col1:
|
121 |
-
st.download_button("CSV", pd.DataFrame(res["papers"]).to_csv(index=False),
|
122 |
-
"papers.csv", "text/csv")
|
123 |
-
with col2:
|
124 |
-
st.download_button("PDF", _pdf(res["papers"]), "papers.pdf", "application/pdf")
|
125 |
-
if st.button("💾 Save"):
|
126 |
-
save_query(st.session_state.last_query, res)
|
127 |
-
st.success("Saved to workspace")
|
128 |
-
|
129 |
-
st.subheader("UMLS concepts")
|
130 |
-
for c in res["umls"]:
|
131 |
-
if c.get("cui"):
|
132 |
-
st.write(f"- **{c['name']}** ({c['cui']})")
|
133 |
-
|
134 |
-
st.subheader("OpenFDA safety")
|
135 |
-
for d in res["drug_safety"]:
|
136 |
-
st.json(d)
|
137 |
-
|
138 |
-
st.subheader("AI summary")
|
139 |
-
st.info(res["ai_summary"])
|
140 |
-
|
141 |
-
with tabs[1]:
|
142 |
-
st.header("Gene / Variant signals")
|
143 |
-
for g in res["genes"]:
|
144 |
-
st.write(f"- **{g.get('name', g.get('geneid'))}** {g.get('description','')}")
|
145 |
-
if res["gene_disease"]:
|
146 |
-
st.markdown("### DisGeNET links")
|
147 |
-
st.json(res["gene_disease"][:15])
|
148 |
-
if res["mesh_defs"]:
|
149 |
-
st.markdown("### MeSH definitions")
|
150 |
-
for d in res["mesh_defs"]:
|
151 |
-
if d:
|
152 |
-
st.write("-", d)
|
153 |
-
|
154 |
-
with tabs[2]:
|
155 |
-
st.header("Clinical trials")
|
156 |
-
if not res["clinical_trials"]:
|
157 |
-
st.info("No trials (rate-limited or none found).")
|
158 |
-
for t in res["clinical_trials"]:
|
159 |
-
st.markdown(f"**{t['NCTId'][0]}** – {t['BriefTitle'][0]}")
|
160 |
-
st.write(f"Phase {t.get('Phase',[''])[0]} | Status {t['OverallStatus'][0]}")
|
161 |
-
|
162 |
-
with tabs[3]:
|
163 |
-
nodes, edges, cfg = build_agraph(res["papers"], res["umls"], res["drug_safety"])
|
164 |
-
hl = st.text_input("Highlight node:", key="hl")
|
165 |
-
if hl:
|
166 |
-
pat = re.compile(re.escape(hl), re.I)
|
167 |
-
for n in nodes:
|
168 |
-
n.color = "#f1c40f" if pat.search(n.label) else "#d3d3d3"
|
169 |
-
agraph(nodes, edges, cfg)
|
170 |
-
|
171 |
-
with tabs[4]:
|
172 |
-
nodes, edges, _ = build_agraph(res["papers"], res["umls"], res["drug_safety"])
|
173 |
-
G = build_nx([n.__dict__ for n in nodes], [e.__dict__ for e in edges])
|
174 |
-
st.metric("Density", f"{get_density(G):.3f}")
|
175 |
-
st.markdown("**Top hubs**")
|
176 |
-
for nid, sc in get_top_hubs(G):
|
177 |
-
lab = next((n.label for n in nodes if n.id == nid), nid)
|
178 |
-
st.write(f"- {lab} {sc:.3f}")
|
179 |
-
|
180 |
-
with tabs[5]:
|
181 |
-
years = [p["published"] for p in res["papers"] if p.get("published")]
|
182 |
-
if years:
|
183 |
-
st.plotly_chart(px.histogram(years, nbins=12, title="Publication Year"))
|
184 |
-
|
185 |
-
# Follow-up Q&A block with callback
|
186 |
-
st.markdown("---")
|
187 |
-
st.text_input("Ask follow‑up question:", key="followup_input")
|
188 |
-
def handle_followup():
|
189 |
-
follow = st.session_state.followup_input
|
190 |
-
if follow.strip():
|
191 |
-
ans = asyncio.run(answer_ai_question(
|
192 |
-
follow,
|
193 |
-
context=st.session_state.last_query,
|
194 |
-
llm=st.session_state.last_llm))
|
195 |
-
st.session_state.followup_response = ans["answer"]
|
196 |
-
else:
|
197 |
-
st.session_state.followup_response = None
|
198 |
-
|
199 |
-
st.button("Ask AI", on_click=handle_followup)
|
200 |
-
|
201 |
-
if st.session_state.followup_response:
|
202 |
-
st.write(st.session_state.followup_response)
|
203 |
-
|
204 |
-
else:
|
205 |
st.info("Enter a question and press **Run Search 🚀**")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
206 |
|
|
|
207 |
if __name__ == "__main__":
|
208 |
render_ui()
|
|
|
1 |
#!/usr/bin/env python3
|
2 |
+
# MedGenesis AI · CPU-only Streamlit front-end (OpenAI / Gemini)
|
3 |
|
4 |
+
from __future__ import annotations
|
5 |
+
import os, pathlib, asyncio, re
|
6 |
from pathlib import Path
|
7 |
|
8 |
import streamlit as st
|
|
|
11 |
from fpdf import FPDF
|
12 |
from streamlit_agraph import agraph
|
13 |
|
14 |
+
from mcp.orchestrator import orchestrate_search, answer_ai_question
|
15 |
+
from mcp.workspace import get_workspace, save_query, clear_workspace
|
16 |
from mcp.knowledge_graph import build_agraph
|
17 |
+
from mcp.graph_utils import build_nx, get_top_hubs, get_density
|
18 |
+
from mcp.alerts import check_alerts
|
19 |
|
20 |
+
# ── Streamlit telemetry dir fix ──────────────────────────────────────
|
21 |
+
os.environ["STREAMLIT_DATA_DIR"] = "/tmp/.streamlit"
|
22 |
+
os.environ["XDG_STATE_HOME"] = "/tmp"
|
23 |
os.environ["STREAMLIT_BROWSER_GATHERUSAGESTATS"] = "false"
|
24 |
pathlib.Path("/tmp/.streamlit").mkdir(parents=True, exist_ok=True)
|
25 |
|
26 |
ROOT = Path(__file__).parent
|
27 |
LOGO = ROOT / "assets" / "logo.png"
|
28 |
|
29 |
+
# -------------------------------------------------------------------#
|
30 |
+
# Utility helpers #
|
31 |
+
# -------------------------------------------------------------------#
|
32 |
def _latin1_safe(txt: str) -> str:
|
33 |
return txt.encode("latin-1", "replace").decode("latin-1")
|
34 |
|
35 |
+
|
36 |
+
def _pdf(papers: list[dict]) -> bytes:
|
37 |
pdf = FPDF()
|
38 |
pdf.set_auto_page_break(auto=True, margin=15)
|
39 |
pdf.add_page()
|
|
|
49 |
pdf.ln(1)
|
50 |
return pdf.output(dest="S").encode("latin-1", "replace")
|
51 |
|
52 |
+
|
53 |
+
def _workspace_sidebar() -> None:
|
54 |
with st.sidebar:
|
55 |
st.header("🗂️ Workspace")
|
56 |
ws = get_workspace()
|
57 |
if not ws:
|
58 |
st.info("Run a search then press **Save** to populate this list.")
|
59 |
return
|
60 |
+
if st.button("Clear workspace 🗑️"):
|
61 |
+
clear_workspace()
|
62 |
+
st.experimental_rerun()
|
63 |
for i, item in enumerate(ws, 1):
|
64 |
with st.expander(f"{i}. {item['query']}"):
|
65 |
st.write(item["result"]["ai_summary"])
|
66 |
|
67 |
+
|
68 |
+
# -------------------------------------------------------------------#
|
69 |
+
# Streamlit main UI #
|
70 |
+
# -------------------------------------------------------------------#
|
71 |
+
def render_ui() -> None:
|
72 |
st.set_page_config("MedGenesis AI", layout="wide")
|
73 |
|
74 |
+
# ── session_state bootstrap ────────────────────────────────────
|
75 |
+
for key, default in {
|
76 |
+
"query_result" : None,
|
77 |
+
"followup_input" : "",
|
78 |
+
"followup_response" : None,
|
79 |
+
"last_query" : "",
|
80 |
+
"last_llm" : "",
|
81 |
+
}.items():
|
82 |
+
st.session_state.setdefault(key, default)
|
|
|
|
|
83 |
|
84 |
_workspace_sidebar()
|
85 |
|
86 |
+
# ── header ─────────────────────────────────────────────────────
|
87 |
c1, c2 = st.columns([0.15, 0.85])
|
88 |
with c1:
|
89 |
if LOGO.exists():
|
|
|
92 |
st.markdown("## 🧬 **MedGenesis AI**")
|
93 |
st.caption("Multi-source biomedical assistant · OpenAI / Gemini")
|
94 |
|
95 |
+
llm = st.radio("LLM engine", ["openai", "gemini"], horizontal=True)
|
96 |
+
query = st.text_input("Enter biomedical question",
|
97 |
+
placeholder="e.g. CRISPR glioblastoma therapy")
|
98 |
|
99 |
+
# ── alerts for saved queries ───────────────────────────────────
|
100 |
+
if ws := get_workspace():
|
101 |
try:
|
102 |
+
news = asyncio.run(check_alerts([w["query"] for w in ws]))
|
103 |
if news:
|
104 |
with st.sidebar:
|
105 |
st.subheader("🔔 New papers")
|
|
|
108 |
except Exception:
|
109 |
pass
|
110 |
|
111 |
+
# ── primary search trigger ─────────────────────────────────────
|
112 |
+
if st.button("Run Search 🚀") and query.strip():
|
113 |
with st.spinner("Collecting literature & biomedical data …"):
|
114 |
res = asyncio.run(orchestrate_search(query, llm=llm))
|
115 |
st.success(f"Completed with **{res['llm_used'].title()}**")
|
116 |
+
st.session_state.update({
|
117 |
+
"query_result" : res,
|
118 |
+
"last_query" : query,
|
119 |
+
"last_llm" : llm,
|
120 |
+
"followup_input" : "",
|
121 |
+
"followup_response" : None,
|
122 |
+
})
|
123 |
|
124 |
res = st.session_state.query_result
|
125 |
+
if not res:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
st.info("Enter a question and press **Run Search 🚀**")
|
127 |
+
return
|
128 |
+
|
129 |
+
# ----------------------------------------------------------------#
|
130 |
+
# Tabs #
|
131 |
+
# ----------------------------------------------------------------#
|
132 |
+
tabs = st.tabs(["Results", "Genes", "Trials", "Graph", "Metrics", "Visuals"])
|
133 |
+
|
134 |
+
# Results ---------------------------------------------------------
|
135 |
+
with tabs[0]:
|
136 |
+
for i, p in enumerate(res["papers"], 1):
|
137 |
+
st.markdown(f"**{i}. [{p['title']}]({p['link']})** *{p['authors']}*")
|
138 |
+
st.write(p["summary"])
|
139 |
+
|
140 |
+
c1, c2 = st.columns(2)
|
141 |
+
with c1:
|
142 |
+
st.download_button("CSV",
|
143 |
+
pd.DataFrame(res["papers"]).to_csv(index=False),
|
144 |
+
"papers.csv", "text/csv")
|
145 |
+
with c2:
|
146 |
+
st.download_button("PDF", _pdf(res["papers"]),
|
147 |
+
"papers.pdf", "application/pdf")
|
148 |
+
|
149 |
+
if st.button("💾 Save this result"):
|
150 |
+
save_query(st.session_state.last_query, res)
|
151 |
+
st.success("Saved to workspace")
|
152 |
+
|
153 |
+
st.subheader("UMLS concepts")
|
154 |
+
for c in res["umls"]:
|
155 |
+
if c.get("cui"):
|
156 |
+
st.write(f"- **{c['name']}** ({c['cui']})")
|
157 |
+
|
158 |
+
st.subheader("OpenFDA safety")
|
159 |
+
for d in res["drug_safety"]:
|
160 |
+
st.json(d)
|
161 |
+
|
162 |
+
st.subheader("AI summary")
|
163 |
+
st.info(res["ai_summary"])
|
164 |
+
|
165 |
+
# Genes -----------------------------------------------------------
|
166 |
+
with tabs[1]:
|
167 |
+
st.header("Gene / Variant signals")
|
168 |
+
if not res["genes"]:
|
169 |
+
st.info("No gene hits (rate-limited or none found).")
|
170 |
+
for g in res["genes"]:
|
171 |
+
st.write(f"- **{g.get('symbol', g.get('name', ''))}** "
|
172 |
+
f"{g.get('summary', '')[:120]}…")
|
173 |
+
|
174 |
+
if res["gene_disease"]:
|
175 |
+
st.markdown("### DisGeNET links")
|
176 |
+
st.json(res["gene_disease"][:15])
|
177 |
+
|
178 |
+
if res["mesh_defs"]:
|
179 |
+
st.markdown("### MeSH definitions")
|
180 |
+
for d in res["mesh_defs"]:
|
181 |
+
if d:
|
182 |
+
st.write("-", d)
|
183 |
+
|
184 |
+
# Trials ----------------------------------------------------------
|
185 |
+
with tabs[2]:
|
186 |
+
st.header("Clinical trials")
|
187 |
+
trials = res["clinical_trials"]
|
188 |
+
if not trials:
|
189 |
+
st.info("No trials (rate-limited or none found).")
|
190 |
+
for t in trials:
|
191 |
+
st.markdown(f"**{t['nctId']}** – {t['briefTitle']}")
|
192 |
+
st.write(f"Phase {t.get('phase','')} | Status {t.get('status')}")
|
193 |
+
|
194 |
+
# Graph -----------------------------------------------------------
|
195 |
+
with tabs[3]:
|
196 |
+
nodes, edges, cfg = build_agraph(
|
197 |
+
res["papers"], res["umls"], res["drug_safety"],
|
198 |
+
res["genes"], res["clinical_trials"], res.get("ot_associations", [])
|
199 |
+
)
|
200 |
+
hl = st.text_input("Highlight node:")
|
201 |
+
if hl:
|
202 |
+
pat = re.compile(re.escape(hl), re.I)
|
203 |
+
for n in nodes:
|
204 |
+
n.color = "#f1c40f" if pat.search(n.label) else "#d3d3d3"
|
205 |
+
agraph(nodes, edges, cfg)
|
206 |
+
|
207 |
+
# Metrics ---------------------------------------------------------
|
208 |
+
with tabs[4]:
|
209 |
+
G = build_nx([n.__dict__ for n in nodes], [e.__dict__ for e in edges])
|
210 |
+
st.metric("Density", f"{get_density(G):.3f}")
|
211 |
+
st.markdown("**Top hubs**")
|
212 |
+
for nid, sc in get_top_hubs(G):
|
213 |
+
lab = next((n.label for n in nodes if n.id == nid), nid)
|
214 |
+
st.write(f"- {lab} {sc:.3f}")
|
215 |
+
|
216 |
+
# Visuals ---------------------------------------------------------
|
217 |
+
with tabs[5]:
|
218 |
+
years = [p["published"] for p in res["papers"] if p.get("published")]
|
219 |
+
if years:
|
220 |
+
st.plotly_chart(px.histogram(years, nbins=12,
|
221 |
+
title="Publication Year"))
|
222 |
+
|
223 |
+
# ----------------------------------------------------------------#
|
224 |
+
# Follow-up Q & A #
|
225 |
+
# ----------------------------------------------------------------#
|
226 |
+
st.markdown("---")
|
227 |
+
st.text_input("Ask follow-up question:",
|
228 |
+
key="followup_input",
|
229 |
+
placeholder="e.g. Any phase III trials recruiting now?")
|
230 |
+
|
231 |
+
def _on_ask() -> None:
|
232 |
+
q = st.session_state.followup_input.strip()
|
233 |
+
if not q:
|
234 |
+
st.warning("Please type a question first.")
|
235 |
+
return
|
236 |
+
with st.spinner("Querying LLM …"):
|
237 |
+
ans = asyncio.run(
|
238 |
+
answer_ai_question(q,
|
239 |
+
context=st.session_state.last_query,
|
240 |
+
llm=st.session_state.last_llm))
|
241 |
+
st.session_state.followup_response = ans["answer"]
|
242 |
+
|
243 |
+
st.button("Ask AI", on_click=_on_ask)
|
244 |
+
if st.session_state.followup_response:
|
245 |
+
st.write(st.session_state.followup_response)
|
246 |
+
|
247 |
|
248 |
+
# -------------------------------------------------------------------#
|
249 |
if __name__ == "__main__":
|
250 |
render_ui()
|