Transforming Genomic Complexity into Clinical Insight with 3D Anatomical Visualizations and Multimodal AI.
OncoPath is a state-of-the-art AI-driven platform designed to predict organ-specific metastatic risks for cancer patients. By integrating longitudinal clinical data with genomic mutation profiles from the MSK-MET dataset, OncoPath provides clinicians and researchers with real-time "What-If" simulations to understand how specific genetic mutations (e.g., TP53, KRAS) influence cancer progression across 21 different anatomical sites.
- 3D Metastatic HUD (Heads-Up Display): A high-fidelity, interactive 3D anatomical viewer that visualizes risk intensity as a dynamic heatmap across the human body.
- Multimodal Fusion Engine: Integrates clinical tabular data, 101-gene mutation profiles, and high-fidelity pathology imaging signals using specialized Vision Encoders (Phikon).
- OncoBot Clinical Assistant: A specialized RAG-based clinical AI assistant restricted to oncological reasoning, providing interpreted insights directly within the dashboard.
- Genomic Prototyping: Real-time mutation toggling with a <50ms latency response from the FastAPI-backed XGBoost inference engine.
- Framework: Next.js 14 (App Router)
- 3D Rendering: Three.js, React Three Fiber, React Three Drei
- Animations: Framer Motion
- Styling: Tailwind CSS
- Interactions: Radix UI components
- API: FastAPI, Uvicorn
- Machine Learning: XGBoost, Scikit-learn
- Interpretability: SHAP
- Vision: PyTorch, Phikon (Pathology Foundation Model)
- Database: Supabase (PostgreSQL)
- LLM Context: Anthropic Claude (Haiku & Sonnet)
.
├── oncopath-next/ # Modern Next.js application (Dashboards & 3D Viewer)
├── scripts/ # Python backend implementation
│ ├── api_service.py # FastAPI server for real-time inference
│ ├── train_iteration_3.py # Automated ML pipeline
│ └── extract_embeddings.py # Vision encoder for multimodal fusion
├── models/ # Serialized XGBoost & Vision artifacts
├── iterations/ # Phase-by-phase project documentation
├── data/ # Genomic (MSK-MET) and clinical datasets
└── README.md
- Node.js: v18 or later
- Python: v3.9 or later
- API Keys: Claude API Key (set in
.env)
# Install Python dependencies
pip install -r requirements.txt
# Start the Risk Simulator API
python scripts/api_service.pycd oncopath-next
# Install dependencies
npm install
# Start the development server
npm run devThe dashboard will be live at http://localhost:3000.
Our models are audited for "Genomic Lift"—measuring the performance improvement when adding genetic data to clinical-only baselines.
- 21 Organ Sites: Specifically optimized models for Brain, Lung, Bone, Liver, etc.
- Interpretability: Exact probability reporting and SHAP force plots for transparent reasoning.
- Validation: Audited against clinical literature (e.g., verifying KRAS influence on Colorectal metastasis patterns).
| Name | Role |
|---|---|
| Jason Seh | Project Architect & ML Engineer |
| Mitchell Eickhoff | Full Stack Developer |
| Konrád Gózon | Frontend & 3D Specialist |
| Rocky Shao | Business Lead |
Distributed under the MIT License. See LICENSE for more information.