Guiding and evaluating molecular modifications of aspirin for preferential COX-2 inhibition using machine learning

Aspirin is a classic non-steroidal anti-inflammatory drug (NSAID) that unfortunately carries an inherent risk of gastrointestinal side effects. This is due to its simultaneous action on both COX-1 and COX-2 enzymes, where the former is associated with adverse reactions and the latter with therapeutic effects. This study begins by constructing a compound database that acts on COX-1 and COX-2, followed by an analysis of the ligand-target binding differences from a macroscopic to microscopic perspective, using 12 types of ADME properties and the X-ray diffraction structures of aspirin-COX-1/2 complexes. Subsequently, we replaced the features with descriptors that describe the molecular topological structure to build predictive models for COX-1/2. Based on the binding differences, we used Chemdraw for real-time molecular modification of aspirin. Our findings suggest that increasing the molecular weight, topological polar surface area (TPSA), and partition coefficient (LogP) of the compound, while also making it more elongated, can enhance the preferential inhibition of COX-2 by aspirin. Therefore, we have modified the original aspirin by introducing carbon chains and hydroxyl groups. We designed 10 structurally modified versions of aspirin and used machine learning and molecular docking to verify their effects on the activity changes of two COX enzymes. The predictive results indicate that the modified aspirin has a stronger inhibitory effect on COX-2 and a weaker effect on COX-1. This provides a reference for the development of COX-2 selective inhibitors based on aspirin and also offers an innovative approach for computer-aided design to eliminate adverse drug reactions.