ML_QSAR_Model

This repository, ML_QSAR_model, contains Python scripts for developing machine learning-based QSAR (Quantitative Structure-Activity Relationship) classification models.

Scripts Overview

1. 1a_generate_ecfp.py
   This script generates Extended-Connectivity Fingerprints (ECFP4) with a diameter of 4 bonds using RDKit, an open-source cheminformatics toolkit.

2. 1b_generate_ecfp.py
   A memory-efficient version of 1a_generate_ecfp.py that can handle multiple SDF files.

3. 2_splitting_with_structural_similarity.py
   This script uses RDKit for splitting the dataset, ensuring stratified sampling based on both structural features and activity labels.

4. 3_rf_hyperparameter_tuning.py
   Utilizes GridSearchCV from Scikit-learn to systematically evaluate a predefined grid of hyperparameter values for Random Forest classifiers. The optimal hyperparameter combination is identified through cross-validation.

5. 4_svm_hyperparameter_tuning.py
   Similar to the previous script, but for Support Vector Machine (SVM) classifiers. It also uses GridSearchCV to find the best hyperparameters.

6. 5_model.py
   This script runs the model using optimized and default parameters, plots the average ROC AUC over cross-validation iterations, computes various evaluation metrics, and organizes the results into directories.

Getting Started

To run these scripts, ensure that:

• The required dependencies, including RDKit and Scikit-learn, are installed.
• The correct data is being read from the appropriate directories.

Properly organizing your data and verifying the input paths will ensure the scripts run smoothly.