Artificial intelligence-guided discovery of marine scalaranes as selective COX-2 inhibitors through machine learning and molecular modelling.

Anti-inflammatory agents are critical for managing chronic conditions, yet traditional non-selective inhibitors often lead to severe gastrointestinal or cardiovascular events. This study presents an integrated artificial intelligence (AI) and molecular modeling pipeline to discover safer alternatives from the Red Sea sponge Hyrtios erectus. We screened a comprehensive library of 8415 COX2-related records using four machine learning algorithms, identifying a random forest model (93% accuracy, 86% MCC) as the most robust predictor. Using this model to screen H. erectus-derived scalarane sesterterpenes, we identified four primary hits. Subsequent molecular docking and 100 ns explicit molecular dynamics simulations revealed that two leads, sesterstatin 7 and 24α-methoxypetrosaspongia C, exhibit superior affinities and thermodynamic stability within the COX2 catalytic site. Notably, these compounds achieve selectivity by targeting the COX2-exclusive hydrophilic side pocket, a feature absent in the COX1 isoform. These findings provide a high-impact foundation for lead optimization, offering a potential socio-economic breakthrough in the development of cost-effective, safer anti-inflammatory therapeutics derived from marine biodiversity.