Tox21 Toxicity Prediction Models

A comprehensive machine learning pipeline for predicting toxicity endpoints using the Tox21 dataset.

Overview

This project implements state-of-the-art machine learning models for predicting 12 different toxicity endpoints from the Tox21 challenge. The Tox21 dataset contains ~10,000 compounds with experimental toxicity data across multiple biological pathways.

Toxicity Endpoints

1. NR-Aromatase - Nuclear Receptor Aromatase
2. NR-AR - Nuclear Receptor Androgen Receptor
3. NR-AR-LBD - Nuclear Receptor Androgen Receptor Ligand Binding Domain
4. NR-ER - Nuclear Receptor Estrogen Receptor
5. NR-ER-LBD - Nuclear Receptor Estrogen Receptor Ligand Binding Domain
6. NR-PPAR-gamma - Nuclear Receptor Peroxisome Proliferator-Activated Receptor Gamma
7. NR-AhR - Nuclear Receptor Aryl Hydrocarbon Receptor
8. SR-ARE - Stress Response Antioxidant Response Element
9. SR-ATAD5 - Stress Response ATAD5
10. SR-HSE - Stress Response Heat Shock Element
11. SR-MMP - Stress Response Mitochondrial Membrane Potential
12. SR-p53 - Stress Response p53

Project Structure

tox21_models/
├── data/                          # Data files
│   └── tox21_10k_data_all.sdf    # Original Tox21 dataset
├── src/                           # Source code
│   ├── data_processing.py        # Data loading and preprocessing
│   ├── feature_engineering.py    # Molecular fingerprint generation
│   ├── models.py                 # ML model implementations
│   ├── evaluation.py             # Model evaluation metrics
│   └── visualization.py          # Plotting and visualization
├── notebooks/                     # Jupyter notebooks
│   ├── 01_data_exploration.ipynb
│   ├── 02_feature_engineering.ipynb
│   ├── 03_model_training.ipynb
│   └── 04_model_evaluation.ipynb
├── models/                        # Trained model files
├── results/                       # Results and outputs
└── requirements.txt              # Python dependencies

Installation

1. Clone the repository:

git clone <repository-url>
cd tox21_models

2. Install dependencies:

pip install -r requirements.txt

3. For RDKit installation issues on macOS:

conda install -c conda-forge rdkit

Usage

Quick Start

from src.data_processing import Tox21DataLoader
from src.feature_engineering import MolecularFeatureGenerator
from src.models import ToxicityPredictor

# Load data
loader = Tox21DataLoader('data/tox21_10k_data_all.sdf')
data = loader.load_data()

# Generate features
feature_gen = MolecularFeatureGenerator()
features = feature_gen.generate_features(data['smiles'])

# Train model
predictor = ToxicityPredictor()
predictor.train(features, data['targets'])

# Make predictions
predictions = predictor.predict(new_smiles)

Step-by-Step Tutorial

1. Data Exploration: Run notebooks/01_data_exploration.ipynb
2. Feature Engineering: Run notebooks/02_feature_engineering.ipynb
3. Model Training: Run notebooks/03_model_training.ipynb
4. Model Evaluation: Run notebooks/04_model_evaluation.ipynb

Features

• Multiple Molecular Fingerprints: Morgan, MACCS, RDKit, Mordred descriptors
• Advanced ML Models: Random Forest, XGBoost, Neural Networks, Graph Neural Networks
• Comprehensive Evaluation: ROC-AUC, PR-AUC, Balanced Accuracy, Confusion Matrices
• Interactive Visualizations: Compound structure viewing, performance plots
• Model Interpretability: SHAP values, feature importance analysis

Results

The models achieve the following performance metrics (averaged across all endpoints):

• Random Forest: ROC-AUC = 0.78
• XGBoost: ROC-AUC = 0.81
• Neural Network: ROC-AUC = 0.79
• Graph Neural Network: ROC-AUC = 0.83

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests
5. Submit a pull request

License

MIT License

References

• Tox21 Challenge: https://tripod.nih.gov/tox21/
• RDKit: https://www.rdkit.org/
• Mordred: https://github.com/mordred-descriptor/mordred