🧬 Genetic Guardrail

AI-Powered Pharmacogenomics Clinical Decision Support System (CDSS)

---

🚀 Overview

Genetic Guardrail is a production-ready backend system that leverages genomic data (VCF files) to provide personalized drug safety recommendations. It bridges the gap between genetic insights and clinical decision-making, enabling safer and more effective prescriptions.

---

❗ Problem Statement

Lack of integration between genomic data and clinical decision-making leads to suboptimal and risky medication choices, resulting in adverse drug reactions and ineffective treatments.

---

💡 Solution

Genetic Guardrail analyzes patient genomic data to:

• Identify enzyme phenotypes (e.g., CYP2D6, CYP2C19)
• Assess drug-specific risks using clinical rules (CPIC-based)
• Provide clear, actionable recommendations
• Ensure reliability with zero-failure architecture

---

🏗️ Architecture

VCF File → Agent 1 (Parser) → Agent 2 (Risk Engine)
         → Agent 3 (Clinical Explainer) → Final Output
                         ↓
                       Cache

---

⚙️ Key Features

🧬 Genomic Analysis

• Parses VCF files efficiently (streaming, large-file optimized)
• Maps variants → enzyme phenotypes

---

⚠️ Risk Engine (Clinical Logic)

• Deterministic CPIC-based drug risk evaluation

• Handles:

  • Poor / Intermediate / Normal / Ultra-Rapid metabolizers
  • Insufficient data scenarios

---

🤖 AI Clinical Explanation

• Generates human-readable medical insights
• Fallback-safe (Gemini optional, rule-based backup)

---

⚡ Performance & Reliability

• Zero-failure architecture (no 500 errors)
• Smart caching system (avoids recomputation)
• Early exit optimization for fast processing
• Timeout-safe AI handling

---

🔐 Authentication & User System

• Google OAuth-based authentication
• User-specific data storage (secure and isolated)

---

📂 File Management

• Persistent VCF storage
• File history retrieval per user
• Reuse previous genomic data for analysis

---

🧪 BioNeMo Integration (Simulation Layer)

• Simulates drug interaction for unknown drugs
• Fallback to AI explanation if unavailable

---

🔌 API Endpoints

Endpoint	Method	Description
/auth/login	GET	Google OAuth login
/auth/callback	GET	OAuth callback handler
/files	GET	Fetch user’s VCF history
/check-prescription	POST	Analyze drug risk (file or file_id)

---

🧪 Example Response

{
  "action": "Avoid",
  "risk_level": "High",
  "clinical_note": "CYP2D6 Poor Metabolizer detected. Codeine may be ineffective and unsafe.",
  "confidence": 0.95
}

---

🛠️ Tech Stack

• Backend: FastAPI (Python)
• Database: SQLite
• Authentication: Google OAuth (Authlib)
• AI Integration: Gemini (optional fallback)
• Simulation: NVIDIA BioNeMo (optional)
• Frontend: React / Flutter (in progress)

---

⚡ Getting Started

1️⃣ Clone Repository

git clone https://github.com/HackIndiaXYZ/hackindia-spark-7-north-region-code-sharks.git
cd backend

---

2️⃣ Install Dependencies

pip install -r requirements.txt

---

3️⃣ Setup Environment Variables

Create .env:

GOOGLE_CLIENT_ID=your_client_id
GOOGLE_CLIENT_SECRET=your_client_secret
BIONEMO_API_KEY=your_key

---

4️⃣ Run Server

uvicorn main:app --reload --port 8000

---

5️⃣ Access API Docs

http://127.0.0.1:8000/docs

---

📱 Mobile Integration

The system can be accessed via:

• Flutter mobile app (USB-connected testing supported)
• React frontend dashboard (for visualization)

---

🔐 Security

• User-specific data isolation
• Secure OAuth authentication
• No exposure of sensitive genomic data

---

🚀 Future Scope

• 📊 Advanced visualization (risk charts, dashboards)
• 📄 Clinical PDF report generation
• 📱 Full-feature mobile app
• 🧠 Enhanced AI-driven decision support
• 🧬 Multi-drug interaction analysis

---

🏆 Why This Project Matters

Genetic Guardrail enables:

• Safer prescriptions
• Personalized medicine
• Reduced adverse drug reactions
• Data-driven clinical decisions

---

👨‍💻 Team

Code Sharks 🦈 HackIndia Spark 7 – North Region

---

📌 Final Note

This project demonstrates a scalable, reliable, and clinically relevant system that brings us closer to the future of precision medicine.

---