Classification of Cyclooxygenase-2 Inhibitors Using Support Vector Machine and Random Forest Methods.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Pharmacology

Background:

• Cyclooxygenase-2 (COX-2) inhibitors are crucial therapeutic targets.
• Developing accurate predictive models for COX-2 inhibition is essential for efficient drug discovery.
• A comprehensive dataset of COX-2 inhibitors is needed for robust model development.

Purpose of the Study:

• To build and validate machine learning classification models for COX-2 inhibitors.
• To identify key structural features contributing to COX-2 inhibitory activity.
• To provide tools for virtual screening of potential COX-2 inhibitor compounds.

Main Methods:

• Utilized a large dataset of 2925 COX-2 inhibitors from 168 literature sources.
• Applied machine learning algorithms: Support Vector Machine (SVM) and Random Forest (RF).
• Developed and evaluated 12 classification models, including analysis of reduced datasets and external test sets using ECFP_4 fingerprints.

Main Results:

• Achieved high performance with best SVM (MCC 0.73) and RF (MCC 0.72) models on the initial dataset.
• New models on a reduced dataset (1630 molecules) yielded MCC values above 0.72.
• Identified critical substructures for activity, including halogen, carboxyl, sulfonamide, and methanesulfonyl groups, and aromatic nitrogen atoms.

Conclusions:

• Machine learning models, particularly RF with ECFP_4 fingerprints, show promise for predicting COX-2 inhibition.
• The identified structural features offer valuable insights for designing novel COX-2 inhibitors.
• These models can significantly aid in prioritizing compounds for experimental validation, accelerating drug discovery efforts.