Machine Learning-Aided Drug Repurposing for Screening COX-2 Inhibitors from Traditional Chinese Medicines.

Area of Science:

• Computational chemistry and pharmacology
• Machine learning applications in drug discovery
• Oncology research

Background:

• Colorectal cancer (CRC) poses a significant global health burden due to high morbidity and mortality.
• Cyclooxygenase-2 (COX-2) is a critical therapeutic target for CRC drug development.
• Machine learning (ML) offers enhanced efficiency and speed in novel drug discovery.

Purpose of the Study:

• To compare the efficacy of Random Forest Classifier (RFC), deep learning (DL), and graph neural network (GNN) models for identifying COX-2 inhibitors.
• To screen Traditional Chinese Medicine (TCM)-derived compounds for potential COX-2 inhibitory activity.
• To identify key molecular substructures responsible for COX-2 inhibition and validate candidate compounds experimentally.

Main Methods:

• Systematic comparison of RFC, DL, and various GNN models (GAT, GCN, MPNN) using diverse features.
• Classification task on a custom COX-2 inhibitor dataset and a TCM-derived compound library.
• Screening of potential COX-2 inhibitors using the optimal model, followed by substructure analysis and experimental validation.

Main Results:

• RFC and DL models demonstrated superior performance over GNN models in COX-2 inhibitor classification.
• The RFC model was identified as the optimal choice for screening TCM-derived compounds.
• Dehydrocostus lactone was experimentally validated as a potent COX-2 inhibitor, with its molecular interactions analyzed.

Conclusions:

• The RFC model is highly effective for screening bioactive compounds from TCM, even with small datasets.
• This study provides a robust methodology for identifying novel drug candidates from natural product libraries.
• The findings support the use of ML in accelerating the development of targeted cancer therapies.