An Explainable 2D-QSAR Machine Learning Approach for Predicting COX-2 Inhibitory Activity Using Molecular...

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery
• Artificial intelligence in medicine

Background:

• Cyclooxygenase-2 (COX-2) is a key target for anti-inflammatory drug development.
• Identifying novel COX-2 inhibitors is crucial for pharmaceutical research.
• Existing methods require enhancement for efficient virtual screening.

Purpose of the Study:

• To develop a robust and interpretable machine learning (ML) framework for predicting COX-2 inhibitory activity.
• To support and accelerate virtual screening efforts in drug discovery.
• To identify key structural features contributing to COX-2 inhibition.

Main Methods:

• A dataset of 2052 compounds from the ChEMBL database was utilized.
• Molecular structures were represented using Morgan fingerprints from SMILES.
• Ensemble ML methods, including Random Forest, were trained and validated.
• SHAP analysis was employed for model interpretability.

Main Results:

• Ensemble methods, particularly Random Forest, showed superior predictive performance.
• Y-randomization tests confirmed the robustness of the model.
• SHAP analysis identified pharmacophore-aligned structural features driving activity.
• A web application for real-time COX-2 activity prediction was developed.

Conclusions:

• Explainable ML effectively predicts COX-2 inhibitory activity.
• The developed framework enhances COX-2 inhibitor discovery.
• The tool facilitates efficient virtual screening in drug development.