TechRx

PharmaGuard

AI-Powered Pharmacogenomic Risk Prediction System

One drug. Two patients. One survives. The difference is in their DNA and PharmaGuard finds it.

RIFT 2026 Hackathon · HealthTech / Pharmacogenomics Track

---

PharmaGuard

Genetics determines how your body processes medication. The same prescription that cures one patient can be toxic or completely useless for another. Yet most clinical workflows still treat all patients the same way.

PharmaGuard changes that. Upload a patient's VCF (Variant Call Format) genomic file and a list of drugs. The system will:

1. Parse the VCF and pull pharmacogenomic variants from 7 clinically critical genes
2. Convert detected star alleles into diplotypes and phenotypes via CPIC guidelines
3. Assign a risk label per drug: Safe · Adjust Dosage · Toxic · Ineffective
4. Write plain-English clinical explanations using GPT-4 (with a rule-based fallback)
5. Package everything into a structured JSON report ready for clinical action

---

Links

🌐 Live Demo	https://techrx.netlify.app/
📹 Demo Video	https://www.linkedin.com/posts/h-r-madalambika-793502368_riftxpwioi-hackathon-24hourchallenge-activity-7430438593418133504-CoKn?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAFszDd8BuKVZ9NmpEL7tDXVnY2y9C3C3W_g

---

Tech Stack

Layer	Technology
Frontend	React 18, Vite, CSS Variables
Backend	FastAPI, Python 3.10+
VCF Parsing	Custom Python parser (VCF v4.2)
Risk Engine	CPIC guideline lookup tables
AI Explanations	OpenAI GPT-4 (rule-based fallback included)
Deployment	Vercel (frontend) + Render (backend)

---

Supported Genes & Drugs

Gene	Drug	Risk if Impaired
CYP2D6	Codeine	Poor metabolizer → Ineffective (no morphine conversion)
CYP2D6	Atomoxetine	Poor metabolizer → Toxic (10× plasma accumulation)
CYP2C19	Clopidogrel	Poor metabolizer → Ineffective + cardiovascular risk
CYP2C9	Warfarin	Poor metabolizer → Toxic (bleeding risk)
SLCO1B1	Simvastatin	*5 variant → Toxic (myopathy, rhabdomyolysis)
TPMT	Azathioprine	Poor metabolizer → Toxic (life-threatening myelosuppression)
DPYD	Fluorouracil	Poor metabolizer → Toxic (fatal multi-organ toxicity)
CYP2B6	Efavirenz	Poor metabolizer → CNS toxicity risk

Phenotypes supported: Poor Metabolizer · Intermediate Metabolizer · Normal Metabolizer · Rapid Metabolizer · Ultrarapid Metabolizer

---

Architecture

┌─────────────────────────────────────────────────────┐
│                    React Frontend                   │
│  FileUpload → DrugInput → Analyze → ResultsDisplay  │
└─────────────────────┬───────────────────────────────┘
                      │ HTTP POST /analyze (multipart)
┌─────────────────────▼───────────────────────────────┐
│                   FastAPI Backend                   │
│                                                     │
│  ┌──────────────┐  ┌─────────────┐  ┌────────────┐  │
│  │  VCF Parser  │→ │ Risk Engine │→ │LLM Explainer│ │
│  └──────────────┘  └─────────────┘  └────────────┘  │
│                                                     │
│  Gene → Diplotype → Phenotype → Risk Label → JSON   │
└─────────────────────────────────────────────────────┘

Data Flow:

VCF File
  → Parse variants & filter by genotype (GT ≠ 0/0)
  → Extract star alleles per gene
  → Infer diplotype
  → Lookup phenotype (CPIC tables)
  → Predict drug risk
  → Generate LLM explanation
  → Return structured JSON report

---

---

Getting Started

Prerequisites

• Python 3.10+
• Node.js 18+
• npm or yarn

Backend Setup

cd backend
pip install -r requirements.txt

# Start server
uvicorn main:app --reload --port 8000

Frontend Setup

cd frontend
npm install

cp .env.example .env

npm run dev

Frontend runs at http://localhost:5173

---

Usage Examples

Basic analysis via curl:

curl -X POST http://localhost:8000/analyze \
  -F "vcf_file=@patient.vcf" \
  -F "drugs=CODEINE,WARFARIN" \
  -F "patient_id=PATIENT_001"

Run the built-in sample demo:

curl -X POST http://localhost:8000/analyze/sample \
  -F "drugs=FLUOROURACIL,AZATHIOPRINE"

Expected VCF file format:

##fileformat=VCFv4.2
##INFO=<ID=GENE,Number=1,Type=String,Description="Gene symbol">
##INFO=<ID=STAR,Number=1,Type=String,Description="Star allele">
##INFO=<ID=RS,Number=1,Type=String,Description="dbSNP rsID">
##FORMAT=<ID=GT,Number=1,Type=String,Description="Genotype">
#CHROM  POS       ID  REF  ALT  QUAL  FILTER  INFO                              FORMAT  SAMPLE
chr22   42522613  .   C    T    .     PASS    GENE=CYP2D6;STAR=*4;RS=rs3892097  GT      1/1

Note: Only variants where the patient carries the alternate allele (GT ≠ 0/0) are processed. Homozygous reference calls are filtered out automatically.

---

API Reference

POST /analyze

Submit a patient VCF file to receive pharmacogenomic drug risk results.

Request (multipart/form-data):

Field	Type	Required	Description
vcf_file	File (.vcf)	✓	Patient VCF file
drugs	String	✓	Comma-separated drug names
patient_id	String	—	Optional patient identifier
openai_api_key	String	—	Optional key for GPT-4 explanations

Response (JSON):

{
  "patient_id": "001",
  "drug": "SIMVASTATIN",
  "timestamp": "2026-02-19T23:58:43.452749Z",
  "risk_assessment": {
    "risk_label": "Safe",
    "confidence_score": 0.9,
    "severity": "none"
  },
  "pharmacogenomic_profile": {
    "primary_gene": "SLCO1B1",
    "diplotype": "*1/*1",
    "phenotype": "Normal Function",
    "detected_variants": []
  },
  "clinical_recommendation": {
    "recommendation": "Standard simvastatin dosing is appropriate. Normal SLCO1B1 function ensures adequate hepatic uptake and clearance. Prescribe desired starting dose per disease-specific guidelines (typically 20–40 mg/day). Routine CK monitoring not required.",
    "cpic_recommendation": "Prescribe desired starting dose per guidelines.",
    "requires_dose_adjustment": false,
    "contraindicated": false
  },
  "llm_generated_explanation": {
    "summary": "Patient carries the *1/*1 diplotype in SLCO1B1, resulting in Normal Function status. For SIMVASTATIN, this translates to a Safe risk assessment with none severity.",
    "mechanism": "SLCO1B1 encodes a hepatic uptake transporter that controls SIMVASTATIN uptake into liver cells. The *1/*1 diplotype impairs this transporter, reducing SIMVASTATIN clearance and increasing systemic exposure with risk of moderate muscle toxicity.",
    "clinical_implications": "This patient is expected to respond normally to standard SIMVASTATIN dosing. No pharmacogenomic-based dose adjustments are necessary.",
    "monitoring": "Routine clinical monitoring per standard of care.",
    "generated_by": "rule-based-fallback",
    "generated_at": "2026-02-19T23:58:43.452734"
  },
  "quality_metrics": {
    "vcf_parsing_success": true,
    "total_variants_parsed": 1,
    "genes_detected": ["CYP2D6"],
    "primary_gene_found": false,
    "explanation_source": "rule-based-fallback"
  }
}

POST /analyze/sample

Run a risk analysis using the built-in high-risk demo VCF — no file upload needed.

GET /drugs

Returns the complete list of supported drug names.

---

☁️ Deployment

Backend → Render

1. Push repository to GitHub
2. Create a new Web Service on render.com
3. Set root directory: backend/
4. Build command: pip install -r requirements.txt
5. Start command: uvicorn main:app --host 0.0.0.0 --port $PORT

Frontend → Vercel

1. Import GitHub repo on vercel.com
2. Set root directory: frontend/
3. Add environment variable: VITE_API_URL=https://your-render-url.onrender.com
4. Deploy

---

Known Limitations

• VCF files must include GENE, STAR, and RS tags in the INFO field
• Complex structural variants and copy number variations are not fully supported
• Diplotype inference relies solely on star alleles detected in the VCF
• GPT-4 explanations require a valid OpenAI API key — rule-based fallback activates otherwise

---

Team — NEXA

Name
Harshita S
H R Madalambika
Kirtisree S
Rakshitha U

BNM Institute of Technology | ECE | Bengaluru

---

PharmaGuard — because the right drug for the average patient isn't always the right drug for your patient.