AI-Based Drug Side-Effect Prediction Using Molecular Fingerprints and Deep Learning

A deep learning system for predicting adverse drug reactions (ADRs) using Extended Connectivity Fingerprints (ECFP) molecular representations and multi-label classification.

Overview

This project develops a machine learning pipeline to predict multiple side effects for drugs based on their molecular structure. It combines:

• Molecular fingerprinting: ECFP-4 representation of drug molecules
• Deep learning: Multi-label classification using PyTorch
• Drug data enrichment: ChEMBL and PubChem integration for SMILES canonicalization
• Web interface: Flask-based drug lookup and prediction UI

Features

• 📊 Multi-label prediction on 300+ side effects
• 🔬 Molecular fingerprint-based features (ECFP-4)
• 🧠 PyTorch neural network with optimized thresholds
• 🌐 Interactive web interface for drug queries
• 📈 Comprehensive performance metrics and visualizations
• 🔄 Data processing pipeline for drug SMILES enrichment

Project Structure

.
├── app.py                          # Flask web application
├── model.py                        # Neural network architecture
├── predict.py                      # Inference module
├── drug_lookup.py                  # Drug database queries
├── generate_metrics.py             # Performance evaluation
├── requirements.txt                # Python dependencies
├── checkpoints/                    # Trained model files
│   ├── best_model.pt              # Best performing model
│   ├── final_model.pt             # Final trained model
│   ├── inference_bundle.joblib    # Pre-processed inference data
│   ├── label_names.json           # Side effect label mapping
│   ├── thresholds.npy             # Optimized prediction thresholds
│   └── training_history.json      # Training metrics
├── Data Processing/               # ETL pipeline scripts
│   ├── smiles_from_chembl.py      # Fetch SMILES from ChEMBL
│   ├── pubchem_for_missing.py     # Backfill missing SMILES
│   └── Outliers_removal.py        # Data cleaning
├── static/                         # Web UI assets (CSS, JS)
├── templates/                      # HTML templates
└── plots/                         # Generated visualizations

Installation

Prerequisites

• Python 3.8+
• pip or conda

Setup

1. Clone the repository

   git clone https://github.com/CharithKalasi/DrugSideEffectPrediction.git
   cd DrugSideEffectPrediction

2. Create a virtual environment (recommended)

   # Windows (PowerShell)
   python -m venv venv
   .\venv\Scripts\Activate.ps1
   
   # Linux/macOS
   python -m venv venv
   source venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Verify setup

   python validate_setup.py

Quick Start

Web Interface

Launch the interactive drug lookup and prediction interface:

python app.py

Access at http://localhost:5000 in your browser.

Command-Line Predictions

Get predictions for a drug:

from predict import predict_side_effects

# Predict side effects for a drug
predictions = predict_side_effects("aspirin")
print(predictions)

View Results & Metrics

Generate and display performance metrics:

python generate_metrics.py
python view_results.py

Data Processing Pipeline

The Data Processing/ folder contains scripts for enriching drug SMILES data:

cd "Data Processing"

# 1. Fetch SMILES from ChEMBL
python smiles_from_chembl.py

# 2. Backfill missing SMILES from PubChem
python pubchem_for_missing.py

# 3. Apply data cleaning corrections
python Outliers_removal.py

Inputs/Outputs:

• drug_names.csv → drug_smiles.csv → drug_smiles_completed.csv
• final_multilabel_ADR_dataset.csv → final_multilabel_ADR_dataset_updated.csv

Model Architecture

The neural network for multi-label classification:

• Input: ECFP-4 fingerprints (2048 features)
• Architecture: 3-layer fully connected network with batch normalization
• Output: 300+ binary predictions (one per side effect)
• Optimization: Binary cross-entropy loss with label smoothing

For details, see model.py.

Dependencies

Package	Version	Purpose
torch	≥2.0.0	Deep learning framework
pandas	≥2.0.0	Data manipulation
numpy	≥1.24.0	Numerical computing
scikit-learn	≥1.3.0	ML utilities & metrics
rdkit	≥2023.3.1	Molecular fingerprinting
flask	(implicit)	Web framework
tqdm	≥4.65.0	Progress bars
joblib	≥1.3.0	Serialization
matplotlib	≥3.7.0	Visualization
seaborn	≥0.12.0	Statistical plots

Performance Metrics

The model achieves strong multi-label classification performance on 300+ side effects:

• Detailed metrics available in plots/detailed_metrics.txt
• Confusion matrices and performance plots in plots/

Usage Examples

Example 1: Predict side effects for a drug

python app.py
# Navigate to http://localhost:5000
# Enter drug name to get predictions

Example 2: Batch predictions

from predict import predict_side_effects
import pandas as pd

drugs = ["aspirin", "ibuprofen", "acetaminophen"]
for drug in drugs:
    results = predict_side_effects(drug)
    print(f"{drug}: {results['top_side_effects']}")

Example 3: Export results

python generate_metrics.py
# Metrics exported to plots/detailed_metrics_table.csv

Configuration

Key parameters in model.py:

• BATCH_SIZE: Training batch size (default: 32)
• LEARNING_RATE: Optimizer learning rate (default: 0.001)
• THRESHOLD: Prediction threshold (optimized per label, stored in checkpoints/thresholds.npy)

Troubleshooting

Issue: RDKit import error

• Solution: Ensure rdkit is installed: pip install rdkit

Issue: Model checkpoint not found

• Solution: Verify checkpoint files exist in checkpoints/ directory

Issue: Web app connection refused

• Solution: Check if port 5000 is available or modify Flask port in app.py

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Submit a pull request with a clear description

License

This project is open source. See LICENSE file for details.

Citation

If you use this project in research, please cite:

@software{DrugSideEffectPrediction2024,
  title={AI-Based Drug Side-Effect Prediction Using Molecular Fingerprints and Deep Learning},
  author={Kalasi, Charith},
  year={2024},
  url={https://github.com/CharithKalasi/DrugSideEffectPrediction}
}

References

• RDKit: Molecular fingerprinting - https://www.rdkit.org/
• ChEMBL: Drug database - https://www.ebi.ac.uk/chembl/
• PubChem: Chemical compound database - https://pubchem.ncbi.nlm.nih.gov/
• PyTorch: Deep learning framework - https://pytorch.org/

Contact

For questions or issues, please open a GitHub issue or contact the maintainer.

---

Last Updated: December 2024
Status: Active Development