title: MCP Res
emoji: 📊
colorFrom: red
colorTo: gray
sdk: docker
sdk_version: 1.46.0
app_file: app.py
pinned: false
short_description: Research
MedGenesis AI MedGenesis AI is a biomedical literature discovery workbench that unifies live data from PubMed, arXiv, MyGene.info, ClinicalTrials.gov v2, DisGeNET, openFDA, Open Targets, DrugCentral, UMLS and more—then lets you explore the evidence in a rich Streamlit interface powered by OpenAI or Gemini LLMs.
┌────────────────────────────────────────────┐ │ Streamlit UI (app.py) │ │ • Results / Genes / Trials / Graph tabs │ │ • PDF / CSV export & follow-up Q&A │ └──────────────┬─────────────────────────────┘ │ async calls ┌──────────────▼─────────────────────────────┐ │ Orchestrator (mcp/orchestrator.py) │ │ • pulls PubMed, arXiv │ │ • keyword extraction (spaCy) │ │ • fans-out to MyGene, CT.gov v2, UMLS… │ │ • merges & summarises with LLM │ └──────────────┬─────────────────────────────┘ │ helpers (mcp/*.py) ▼ ┌────────────────────────────────────────────┐ │ External APIs + local TSV (DrugBank) │ └────────────────────────────────────────────┘
🔑 Features Domain Source / API What you get Literature PubMed + arXiv titles, abstracts, authors, year Gene info MyGene.info + NCBI Gene symbol, name, GO, ClinVar, MeSH definitions Trials ClinicalTrials.gov v2 NCT ID, phase, status, start date Disease ↔ gene DisGeNET top associations & scores Drug safety openFDA, DrugCentral adverse events, approvals, MoA Graph edges Open Targets GraphQL gene–disease-drug links (+ OT score) Ontology UMLS, HPO, Wikidata concept CUI, phenotype look-ups
🚀 Quick start bash Copy Edit
clone repo
git clone https://github.com/your-org/medgenesis.git cd medgenesis
build & run locally
python -m venv .venv && source .venv/bin/activate pip install -r requirements.txt python -m spacy download en_core_web_sm streamlit run app.py app.py starts a Streamlit server on localhost:8501. Enter a biomedical question (e.g. “CRISPR glioblastoma therapy”) and press Run Search 🚀.
🐳 Docker / Hugging Face Space The included Dockerfile is CPU-only and downloads the spaCy model at build time:
bash Copy Edit docker build -t medgenesis . docker run -p 7860:7860 -e OPENAI_API_KEY=sk-... medgenesis HF Spaces: push the repo, set the environment secrets below, and Spaces will pick up Dockerfile.
🔐 Environment variables Variable Description OPENAI_API_KEY OpenAI account key (GPT-4o, GPT-4o-mini …) GEMINI_KEY Google Generative AI key (Gemini 1.5 Flash) UMLS_KEY UMLS Licensing key (ticket auth) DISGENET_KEY DisGeNET Bearer token (optional) PUB_KEY NCBI E-utils key (optional, boosts quota) BIO_KEY NCBI E-utils key for Gene/MeSH (optional)
Set them in .env, your shell, or HF Secrets.
🗃️ Local data mcp/data/drugbank_open_structured_drug_links.tsv – DrugBank Open Data Download from the DrugBank Open-Data page and place it here.
The file is lazy-loaded and cached; the app still works without it.
🧪 Tests bash Copy Edit pytest tests/ Unit tests mock external APIs and verify parsing, caching and orchestrator merges.
🛠️ Contributing Fork & create a feature branch.
Follow Conventional Commits for PR titles.
Run pre-commit install to auto-format with black & ruff.
Submit a PR; GitHub Actions will run lint + tests.
📄 License Apache 2.0 – free for research and commercial use. API terms of each external provider still apply.
Happy discovering!