Binary classification of drug molecules using machine learning as a drug repurposing tool for finding new COX-۲ inhibitors

Background and Aims: To accelerate the drug discovery process, virtual screening methods are currently employed as a high-throughput strategy for drug repurposing. The therapeutic target investigated in this study for drug repurposing is the enzyme Cyclooxygenase-۲ (COX-۲). This enzyme is crucial in converting arachidonic acid to prostaglandins, which are involved in inflammatory processes. Beyond its role in inflammation, COX-۲ is frequently overexpressed in cancers (e.g., cervical cancer) and is associated with inhibiting apoptosis and promoting angiogenesis. Consequently, compounds that inhibit this enzyme are commonly used as anti-inflammatory agents. Methods: In this study, molecules tested in vitro against the COX-۲ enzyme were extracted from the ChEMBL database. Subsequently, several machine learning models—including K-Nearest Neighbors (KNN), Support Vector Machines (SVM), XGBoost, CatBoost, LightGBM, Random Forest (RF), and Artificial Neural Networks (ANN)—were trained for binary classification of the molecules based on their inhibitory or non-inhibitory activity against COX-۲. To optimize model performance, multiple feature sets, such as MACCS keys, topological torsion, atom pairs, and RDKit descriptors, were evaluated. Finally, the trained model was applied to screen FDA-approved, experimental, and investigational drugs to identify potential COX-۲ inhibitors. Results: Among FDA-approved and investigational drugs Lipoic Acid, MK-۸۸۶, and Pemirolast can be used for further investigation as the candidates for drug repurposing against COX-۲. The model's predictions can be further validated through molecular docking as an in silico validation step. The proposed candidates can subsequently undergo in vitro testing for experimental validation. Conclusion: The process of discovering new drugs is time-consuming and costly. Therefore, utilizing virtual screening methods assisted by machine learning can significantly contribute to shortening this process by repurposing existing drugs (either FDA-approved or in experimental stages) for new therapeutic purposes.

کلیدواژه ها:

Drug repurposing ، Machine learning ، Cyclooxygenase-۲