💊 Drug Discovery Triage

Production-ready web application for preliminary drug candidate screening with comprehensive ADMET predictions

Live Demo • Documentation • Tech Stack • Features

---

🎯 Overview

Drug Discovery Triage is a modern web application that helps medicinal chemists and researchers rapidly assess drug-likeness and ADMET properties of small molecules. Built with transparency and scientific rigor in mind, it uses validated rule-based methods to predict:

• Drug-likeness scores (QED, Lipinski, Lead-likeness)
• ADMET properties (Solubility, BBB penetration, GI absorption, CNS MPO)
• Toxicity alerts (10 PAINS patterns, CYP450 metabolism)
• Stereochemistry analysis with educational context

🌟 Why This Project Stands Out

• ✅ Production-ready: Full-stack deployment with CI/CD
• ✅ Scientifically validated: All methods peer-reviewed (ESOL, Lipinski, CNS MPO, Murcko scaffolds)
• ✅ Transparent: Rule-based predictions (no black-box ML)
• ✅ Modern UI/UX: Dark-themed Material-UI with responsive design
• ✅ Fast: <100ms prediction time, instant results
• ✅ Educational: Explains WHY stereoisomers differ in biology

---

🚀 Live Demo

Try it now: https://drug-discovery-triage-hx7j.vercel.app

Example Molecules to Test:

Molecule	SMILES	Expected Results
(S)-Ibuprofen	CC(C)Cc1ccc([C@H](C)C(=O)O)cc1	QED: 0.72, BBB+, High CNS
Aspirin	CC(=O)OC1=CC=CC=C1C(=O)O	QED: 0.58, Lead-like
Caffeine	CN1C=NC2=C1C(=O)N(C(=O)N2C)C	BBB+, CNS active

---

✨ Key Features

🧪 Comprehensive ADMET Predictions

Absorption & Distribution Aqueous solubility (ESOL equation) GI absorption (Lipinski + Veber) BBB penetration (multi-parameter) CNS MPO score (Pfizer method)

Metabolism & Toxicity CYP450 substrate predictions 10 PAINS/structural alerts Clinical toxicity risk FDA approval likelihood

📊 Drug-likeness Metrics

• QED Score: Quantitative Estimate of Drug-likeness (0-1 scale)
• Lipinski Rule-of-Five: Oral bioavailability assessment
• Lead-likeness: Rule of 3 for fragment screening
• Scaffold Analysis: Murcko scaffold extraction

🔬 Stereochemistry Analysis

• Identifies chiral centers (R/S configuration)
• Educational context explaining enantiomer differences
• Visual structure comparison
• Historical examples (Thalidomide, Ibuprofen)

🎨 Modern User Interface

• Dark-themed Material-UI design
• Real-time predictions (<100ms)
• Interactive property cards with tooltips
• Responsive layout (mobile-friendly)
• Copy SMILES functionality

---

🛠️ Tech Stack

Backend

• FastAPI - High-performance Python API framework
• RDKit - Cheminformatics toolkit for molecular descriptors
• Pydantic - Data validation and settings management
• Uvicorn - ASGI server

Frontend

• React 19 - Modern UI library
• TypeScript - Type-safe JavaScript
• Material-UI 7 - Component library
• Vite - Fast build tool

Deployment

• Vercel - Frontend hosting with edge functions
• Railway/Render - Backend API hosting
• Docker - Containerization for reproducibility
• GitHub Actions - CI/CD pipeline

---

📚 Documentation

Quick Start

# Clone the repository
git clone https://github.com/mondalsou/drug-discovery-triage.git
cd drug-discovery-triage

# Run with Docker Compose (easiest)
docker-compose up --build

# Access the application
# Frontend: http://localhost:3000
# Backend API: http://localhost:8000
# API Docs: http://localhost:8000/docs

Local Development

Backend:

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload

Frontend:

cd frontend
npm install
npm run dev

---

🔬 Scientific Validation

All prediction methods are based on peer-reviewed literature:

Method	Reference	Year
ESOL Solubility	Delaney, J. Chem. Inf. Comput. Sci.	2004
Lipinski RO5	Lipinski et al., Adv. Drug Deliv. Rev.	2001
CNS MPO	Wager et al., ACS Chem. Neurosci.	2010
Lead-likeness	Congreve et al., Drug Discovery Today	2003
Murcko Scaffolds	Bemis & Murcko, J. Med. Chem.	1996
PAINS	Baell & Holloway, J. Med. Chem.	2010

---

📊 Architecture

┌─────────────────────────────────────────────────────────┐
│                     User Browser                        │
└────────────────────┬────────────────────────────────────┘
                     │
                     ▼
┌─────────────────────────────────────────────────────────┐
│              Frontend (React + Vite)                    │
│              Hosted on Vercel                           │
│  • Material-UI components                               │
│  • TypeScript for type safety                           │
│  • Responsive design                                    │
└────────────────────┬────────────────────────────────────┘
                     │ REST API
                     ▼
┌─────────────────────────────────────────────────────────┐
│            Backend (FastAPI + RDKit)                    │
│            Hosted on Railway/Render                     │
│  • Molecular descriptor calculation                     │
│  • ADMET predictions                                    │
│  • Structure rendering                                  │
└─────────────────────────────────────────────────────────┘

---

🎯 Use Cases

For Medicinal Chemists

• Rapid compound prioritization
• SAR analysis with scaffold extraction
• ADMET liability identification
• Stereochemistry education

For Researchers

• Virtual screening preparation
• Compound library profiling
• Lead optimization guidance
• Drug-likeness assessment

For Educators

• Teaching medicinal chemistry concepts
• Demonstrating ADMET principles
• Stereochemistry visualization
• Property-based design

---

🚢 Deployment

Docker Compose (Self-Hosted)

docker-compose up --build

Vercel (Frontend)

cd frontend
vercel --prod

Railway (Backend)

cd backend
railway up

See DEPLOYMENT_GUIDE.md for detailed instructions.

---

🤝 Contributing

Contributions are welcome! This project is designed to be:

• Educational: Clear, well-documented code
• Extensible: Easy to add new prediction methods
• Scientific: All methods must be peer-reviewed

Areas for Enhancement

• Batch processing (CSV upload)
• Export results (PDF/CSV)
• Additional ADMET endpoints
• 3D conformation generation
• Substructure search

---

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

👨‍💻 Author

Sourav Mondal

• GitHub: @mondalsou
• LinkedIn: Sourav Mondal

---

🙏 Acknowledgments

• RDKit community for the excellent cheminformatics toolkit
• FastAPI for the modern Python web framework
• Material-UI for the beautiful component library
• Scientific literature authors for validated ADMET methods

---

📈 Project Stats

• Lines of Code: ~3,500 (Python + TypeScript)
• Prediction Time: <100ms per molecule
• Docker Image Size: ~500MB (optimized)
• Test Coverage: Core chemistry functions validated
• Deployment: Automated CI/CD with GitHub Actions

---

⭐ Star this repo if you find it useful!

Built with ❤️ for the drug discovery community

Live Demo • Report Bug • Request Feature