BioReason

India's Biomedical Intelligence Platform

Multi-hop reasoning over a 4.3M-edge knowledge graph — extended with Indian population genomics, 17,967 Ayurvedic compounds, and 180+ active Indian clinical trials.

Live Demo · Documentation · Architecture · Contributing

---

Why BioReason

Global biomedical AI is calibrated on Western data. India has different genetics, different prevalent diseases, and a 5,000-year traditional medicine system that no global knowledge graph captures.

Gap	Reality	Why it matters
Population genetics	CYP2C19*2 LoF: 23% in S. Asians vs 15% globally	Standard clopidogrel dosing is wrong for ~300M Indians
Traditional medicine	17,967 IMPPAT phytochemicals → 0 in any global KG	CDSCO needs computational mechanism evidence — no tool provides it
Image analysis	India: 3,000 pathologists for 1.4B people	Retinal cameras everywhere, but no tool connects image → KG → treatment

BioReason is the missing layer.

What it does

Ask a biomedical question in plain English. The system:

1. LLM generates Cypher — Llama 3.3 70B writes a multi-step Neo4j query plan
2. Graph traverses — across 4.3M curated relationships (drugs ↔ proteins ↔ pathways ↔ diseases)
3. India overlay — phytochemical hits, IndiGen variant frequencies, active trials
4. Evidence-graded answer — with confidence ratings, source citations, exportable PDF

Or upload a biomedical image (retinal fundus, blood smear, histopathology, cytology). Vision AI extracts biomarkers → Image-to-KG bridge maps them to graph nodes → treatment paths returned with Indian PGx warnings.

Modules

• Query — natural language multi-hop reasoning
• Drug Repurposing Scanner — find FDA-approved candidates for any disease via shared targets
• Ayurvedic Validation Engine — mechanism certificates for IMPPAT compounds
• Indian Pharmacogenomics Explorer — IndiGen-calibrated drug-gene interactions
• BioReason Vision — image → biomarkers → KG → treatment paths
• Hypothesis Builder — connect any two biomedical entities
• Synergy Explorer — 2.67M curated drug combination edges
• PGx Safety Alerts — 7 high-impact India-specific dosing warnings
• Batch / Compare / Search / Graph stats — supporting tools

Architecture

┌──────────────────┐    ┌──────────────────┐    ┌──────────────────┐
│  Next.js 14      │───▶│  FastAPI         │───▶│  Neo4j Community │
│  (Tailwind, D3)  │    │  (multi-LLM)     │    │  (4.3M edges)    │
└──────────────────┘    └──────────────────┘    └──────────────────┘
        │                       │                        │
        │                       ▼                        │
        │              ┌──────────────────┐              │
        │              │  LLM Provider    │              │
        │              │  Groq / Claude / │              │
        │              │  Ollama / OpenAI │              │
        │              └──────────────────┘              │
        ▼                                                ▼
┌──────────────────┐                         ┌──────────────────┐
│  jsPDF export    │                         │  Pipelines:      │
│  D3 force graph  │                         │  PrimeKG, IMPPAT │
│  PGx alerts UI   │                         │  IndiGen, CTRI   │
└──────────────────┘                         └──────────────────┘

Reasoning pattern: ESCARGOT-style — LLM generates Cypher → Neo4j executes → LLM synthesises results. See api/reason.py.

Vision pattern: Image → multimodal LLM extracts biomarkers → biomarker-to-KG bridge → multi-hop reasoning. See api/vision.py.

Data sources

All open and properly licensed. Loading scripts in pipeline/:

• PrimeKG (Harvard MIMS) — 90K nodes, 4.05M edges
• IMPPAT 2.0 (ACTREC) — 17,967 phytochemicals
• IndiGen (CSIR-IGIB) — Indian PGx variants
• ClinicalTrials.gov — 180 India-specific trials
• DrugBank, UniProt, Reactome, KEGG, OMIM, MONDO, HPO, PharmGKB, GO, Uberon, NCBI Gene

Getting started

Prerequisites

• Python 3.11+
• Node.js 18+
• Neo4j Community 5.x (or Docker)
• A free Groq API key (or Anthropic / Ollama)

One-time setup

git clone https://github.com/shailesh2790/bioreason-india.git
cd bioreason-india

# Backend
pip install -r requirements.txt

# Frontend
npm install

# Environment
cp .env.example .env
# Edit .env: add GROQ_API_KEY, set NEO4J_URI/USER/PASSWORD

Run Neo4j

docker compose up -d   # uses docker-compose.yml — Neo4j 5.x + APOC, port 7687

Load data (one-time, takes time)

python -m pipeline.load_primekg          # ~4 hours, 4.05M edges
python -m pipeline.load_imppat           # ~30 sec, 10 sample compounds
python -m pipeline.load_indigen          # ~5 sec, 14 PGx variants
python -m pipeline.load_clinical_trials  # ~2 min, 180 Indian trials

Start the stack

# Terminal 1 — FastAPI
uvicorn api.reason:app --reload --port 8000

# Terminal 2 — Next.js
npm run dev

Open http://localhost:3000

Configuration

Set in .env:

# LLM provider (groq | anthropic | ollama | together | openrouter)
LLM_PROVIDER=groq
GROQ_API_KEY=gsk_...
GROQ_MODEL=llama-3.3-70b-versatile

# Neo4j
NEO4J_URI=bolt://127.0.0.1:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=bioreason123

# Frontend → Backend
FASTAPI_URL=http://localhost:8000

Production deployment

• Frontend: deploys to Vercel — vercel --prod
• Backend: Dockerfile and railway.toml included for Railway
• Tunnel during demos: see scripts/bioreason-tunnel.ps1 — Cloudflare quick tunnel with auto-restart and Vercel URL sync

API

# Multi-hop reasoning
POST /reason       { "question": "...", "max_hops": 3, "india_context": true }

# Direct Cypher (read-only)
POST /cypher       { "cypher": "MATCH ..." }

# Vision analysis
POST /vision/analyse  (multipart: image, modality, clinical_context)

# Health & stats
GET /health
GET /stats

Full reference: visit /api-docs in the running app.

Roadmap

• ROBOKOP integration (140M edges)
• UNI 2 / RETFound for production-grade vision
• DiffDock molecular docking scores for all 17,967 phytochemicals
• Full IndiGen variant set (~10,000 PGx variants)
• Clinician feedback loop → confidence weight updates
• CDSCO mechanism certificate templates

Contributing

We welcome contributions! See CONTRIBUTING.md.

Good first issues:

• Add a new biomedical data pipeline (e.g. MalaCards, OpenTargets)
• Improve the Image-to-KG biomarker mappings
• Add unit tests
• Translate UI to regional Indian languages

License

Apache License 2.0 — free for academic, research, and commercial use.

The code is open source. The hosted BioReason Cloud service, curated India-specific datasets, and regulatory submission tooling are commercial offerings of the maintainers.

Citation

If you use BioReason in research, please cite:

@software{bioreason2026,
  author = {Tripathi, Shailesh Kumar},
  title  = {BioReason: India's Biomedical Intelligence Platform},
  year   = {2026},
  url    = {https://github.com/shailesh2790/bioreason-india}
}

Acknowledgements

Built on the shoulders of giants — PrimeKG (Harvard MIMS), IMPPAT 2.0 (ACTREC Mumbai), IndiGen (CSIR-IGIB), GenomeIndia (DBT), and the entire Neo4j and Llama open-source ecosystem.

Reasoning pipeline inspired by ESCARGOT (Aug 2024).

---

Made in India 🇮🇳 for 1.4 billion people.

Live Demo · Issues · Discussions