Conserved hydrophobic gatekeeper mechanisms in COX-1-like pockets revealed by AI-driven pocket similarity and molecular dynamics simulations of mofezolac.

Cyclooxygenase-1 (COX-1, PTGS1) has recently re-emerged as a significant therapeutic target in inflammatory diseases and ovarian cancer, particularly due to its high expression in high-grade serous ovarian carcinoma. Elucidating the structural determinants that govern the selective binding of inhibitors within the COX-1 catalytic pocket is essential for rational drug design. In this study, we investigated the binding behavior and potential cross-target activity of the selective COX-1 inhibitor mofezolac using an integrated artificial intelligence-assisted pocket-based drug discovery framework. Dynamic protein binding site's volumetric topology based binding pocket similarity analysis was performed to identify proteins possessing similar binding cavities structurally comparable to the COX-1-mofezolac complex (5WBE). Five structurally related proteins were identified based on conserved dynamic pocket geometry and physicochemical descriptors. Molecular docking and 200 ns MD simulations were subsequently executed to evaluate the stability, interaction dynamics, and binding mechanisms of mofezolac within these COX-1-like pockets. Binding free-energy calculations using the MM/GBSA approach revealed favorable and thermodynamically stable interactions across all systems, with ΔG_bind values ranging from -24.16 to -54.18 kcal/mol. Molecular dynamics analysis further identified conserved phenylalanine residues acting as dynamic hydrophobic gatekeepers, stabilizing mofezolac through persistent aromatic interactions. Computational mutagenesis (Phe→Ala) supported the contribution of these residues to ligand stabilization, as mutation reduced the predicted binding strength. These conserved hydrophobic pocket architectures can facilitate cross-target ligand recognition and highlight the importance of aromatic gatekeeper residues in stabilizing COX-1 inhibitors. Importantly, these findings suggest that mofezolac may engage structurally conserved binding environments across multiple proteins, supporting its potential for multi-target activity and drug repurposing. The integration of these computational methods provides mechanistic insight into mofezolac binding and establishes a machine learning-guided framework for identifying conserved druggable pockets relevant to anti-inflammatory and cancer therapeutic development.