#!/usr/bin/env python3 # MedGenesis AI · CPU-only Streamlit app (OpenAI / Gemini) # ── Streamlit telemetry dir fix ─────────────────────────────────────── import os, pathlib os.environ["STREAMLIT_DATA_DIR"] = "/tmp/.streamlit" os.environ["XDG_STATE_HOME"] = "/tmp" os.environ["STREAMLIT_BROWSER_GATHERUSAGESTATS"] = "false" pathlib.Path("/tmp/.streamlit").mkdir(parents=True, exist_ok=True) # ── Std-lib / third-party imports ──────────────────────────────────── import asyncio, re from pathlib import Path import streamlit as st import pandas as pd import plotly.express as px from fpdf import FPDF # classic FPDF → Latin-1 only from streamlit_agraph import agraph # ── Internal helpers ──────────────────────────────────────────────── from mcp.orchestrator import orchestrate_search, answer_ai_question from mcp.workspace import get_workspace, save_query from mcp.knowledge_graph import build_agraph from mcp.graph_metrics import build_nx, get_top_hubs, get_density from mcp.alerts import check_alerts ROOT = Path(__file__).parent LOGO = ROOT / "assets" / "logo.png" # ── PDF export helper (UTF-8 → Latin-1 “safe”) ────────────────────── def _latin1_safe(txt: str) -> str: """Return text that FPDF(latin-1) can embed; replace unknown chars.""" return txt.encode("latin-1", "replace").decode("latin-1") def _pdf(papers): pdf = FPDF() pdf.set_auto_page_break(auto=True, margin=15) pdf.add_page() pdf.set_font("Helvetica", size=11) pdf.cell(200, 8, _latin1_safe("MedGenesis AI – Results"), ln=True, align="C") pdf.ln(3) for i, p in enumerate(papers, 1): pdf.set_font("Helvetica", "B", 11) pdf.multi_cell(0, 7, _latin1_safe(f"{i}. {p['title']}")) pdf.set_font("Helvetica", "", 9) body = ( f"{p['authors']}\n" f"{p['summary']}\n" f"{p['link']}\n" ) pdf.multi_cell(0, 6, _latin1_safe(body)) pdf.ln(1) return pdf.output(dest="S").encode("latin-1", "replace") # ── Sidebar workspace ─────────────────────────────────────────────── def _workspace_sidebar(): with st.sidebar: st.header("🗂️ Workspace") ws = get_workspace() if not ws: st.info("Run a search then press **Save** to populate this list.") return for i, item in enumerate(ws, 1): with st.expander(f"{i}. {item['query']}"): st.write(item["result"]["ai_summary"]) # ── Main UI ───────────────────────────────────────────────────────── def render_ui(): st.set_page_config("MedGenesis AI", layout="wide") _workspace_sidebar() # Header c1, c2 = st.columns([0.15, 0.85]) with c1: if LOGO.exists(): st.image(str(LOGO), width=105) with c2: st.markdown("## 🧬 **MedGenesis AI**") st.caption("Multi-source biomedical assistant · OpenAI / Gemini") llm = st.radio("LLM engine", ["openai", "gemini"], horizontal=True) query = st.text_input("Enter biomedical question", placeholder="e.g. CRISPR glioblastoma therapy") # Alert check if get_workspace(): try: news = asyncio.run(check_alerts([w["query"] for w in get_workspace()])) if news: with st.sidebar: st.subheader("🔔 New papers") for q, lnks in news.items(): st.write(f"**{q}** – {len(lnks)} new") except Exception: pass # Run search if st.button("Run Search 🚀") and query: with st.spinner("Collecting literature & biomedical data …"): res = asyncio.run(orchestrate_search(query, llm=llm)) st.success(f"Completed with **{res['llm_used'].title()}**") tabs = st.tabs(["Results", "Genes", "Trials", "Graph", "Metrics", "Visuals"]) # Results with tabs[0]: for i, p in enumerate(res["papers"], 1): st.markdown(f"**{i}. [{p['title']}]({p['link']})** *{p['authors']}*") st.write(p["summary"]) col1, col2 = st.columns(2) with col1: st.download_button("CSV", pd.DataFrame(res["papers"]).to_csv(index=False), "papers.csv", "text/csv") with col2: st.download_button("PDF", _pdf(res["papers"]), "papers.pdf", "application/pdf") if st.button("💾 Save"): save_query(query, res) st.success("Saved to workspace") st.subheader("UMLS concepts") for c in res["umls"]: if c.get("cui"): st.write(f"- **{c['name']}** ({c['cui']})") st.subheader("OpenFDA safety") for d in res["drug_safety"]: st.json(d) st.subheader("AI summary") st.info(res["ai_summary"]) # Genes with tabs[1]: st.header("Gene / Variant signals") for g in res["genes"]: st.write(f"- **{g.get('name', g.get('geneid'))}** " f"{g.get('description', '')}") if res["gene_disease"]: st.markdown("### DisGeNET links") st.json(res["gene_disease"][:15]) if res["mesh_defs"]: st.markdown("### MeSH definitions") for d in res["mesh_defs"]: if d: st.write("-", d) # Trials with tabs[2]: st.header("Clinical trials") if not res["clinical_trials"]: st.info("No trials (rate-limited or none found).") for t in res["clinical_trials"]: st.markdown(f"**{t['NCTId'][0]}** – {t['BriefTitle'][0]}") st.write(f"Phase {t.get('Phase', [''])[0]} " f"| Status {t['OverallStatus'][0]}") # Graph with tabs[3]: nodes, edges, cfg = build_agraph(res["papers"], res["umls"], res["drug_safety"]) hl = st.text_input("Highlight node:", key="hl") if hl: pat = re.compile(re.escape(hl), re.I) for n in nodes: n.color = "#f1c40f" if pat.search(n.label) else "#d3d3d3" agraph(nodes, edges, cfg) # Metrics with tabs[4]: G = build_nx([n.__dict__ for n in nodes], [e.__dict__ for e in edges]) st.metric("Density", f"{get_density(G):.3f}") st.markdown("**Top hubs**") for nid, sc in get_top_hubs(G): lab = next((n.label for n in nodes if n.id == nid), nid) st.write(f"- {lab} {sc:.3f}") # Visuals with tabs[5]: years = [p["published"] for p in res["papers"] if p.get("published")] if years: st.plotly_chart(px.histogram(years, nbins=12, title="Publication Year")) # Follow-up Q-A st.markdown("---") follow = st.text_input("Ask follow-up:") if st.button("Ask AI"): ans = asyncio.run(answer_ai_question(follow, context=query, llm=llm)) st.write(ans["answer"]) else: st.info("Enter a question and press **Run Search 🚀**") # entry-point if __name__ == "__main__": render_ui()