🧪 Toxicity Predictor (SR-MMP)

This project is a lightweight machine learning tool that predicts molecular toxicity using the SR-MMP (mitochondrial membrane potential) endpoint from the Tox21 dataset.

---

🚀 Features

• Parses SMILES strings from real chemical data
• Calculates Mordred molecular descriptors
• Trains a Random Forest model for binary toxicity classification
• Evaluates model performance with precision, recall, F1-score, and ROC AUC
• Predicts toxicity on new molecules via SMILES input
• Provides exploratory data analysis and molecule visualization

---

🧠 Technologies

• Python 3.x
• RDKit
• Mordred
• Scikit-learn
• Seaborn / Matplotlib
• Pandas / NumPy
• Jupyter Notebooks

---

📂 Project Structure

toxicity-predictor/
├── data/                      # Cleaned dataset and input/output molecules
│   └── new_molecules.csv
│   └── predictions.csv
│   └── tox21_sr-mmp.csv
|   └── tox21.csv.gz (which has been preprocessed by Step_1_preprocess.py to tox21_sr-mmp)
│
├── scripts/                   # Python scripts for pipeline stages
│   ├── Step_1_preprocess.py   # Clean original dataset
│   ├── Step_2_train_model.py  # Train and evaluate RandomForest model
│   └── Step_3_tox_predict.py  # Predict toxicity for new SMILES
│
├── model/                     # Trained model + feature names
│   ├── descriptor_names.txt
│   └── random_forest_model.pkl
│
├── notebooks/                 # Jupyter notebooks for EDA
│   └── eda.ipynb
│
├── .gitattributes
├── .gitignore                 # Files/folders to ignore
├── README.md                  # Project overview
└── requirements.txt               # Python dependencies

---

🧪 Exploratory Data Analysis

All EDA is located in notebooks/eda.ipynb. It includes:

• Toxicity class imbalance visualization
• Distribution of molecular properties (Molecular Weight, LogP, TPSA, Rotatable Bonds)
• Paired visual comparison of toxic vs non-toxic molecules based on structural similarity
• Top 20 feature importances from trained Random Forest model

---

🏁 Getting Started

1. Clone the repository

git clone https://github.com/YOUR_USERNAME/toxicity-predictor.git
cd toxicity-predictor

2. Set up a virtual environment (optional but recommended)

python -m venv venv
source venv/bin/activate         # On Windows: .\venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Run the model training script

python scripts/train_model.py

5. Predict toxicity on new molecules

python scripts/predict_toxicity.py

---

📊 Example output of model training

              precision    recall  f1-score   support

           0       0.86      0.99      0.92       121
           1       0.89      0.29      0.43        28

    accuracy                           0.86       149
   macro avg       0.87      0.64      0.68       149
weighted avg       0.86      0.86      0.83       149

ROC AUC Score: 0.899

---

📊 Example output of predicting the toxicity of new molecules

smiles,mol,toxicity_prediction
CC(=O)OC1=CC=CC=C1C(=O)O,<rdkit.Chem.rdchem.Mol object at 0x00000234A8498F90>,0
CN1CCC(CC1)NC2=NC=NC3=CC=CC=C23,<rdkit.Chem.rdchem.Mol object at 0x00000234B8729690>,0
CC(C)CC(=O)O,<rdkit.Chem.rdchem.Mol object at 0x00000234B8729620>,0
CCOc1ccc(cc1OC)C(=O)O,<rdkit.Chem.rdchem.Mol object at 0x00000234B8729700>,0

---

📜 License

This project is open source under the MIT License.

---

🤝 Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you’d like to change or improve.