Artificial Intelligence- and Machine Learning-Assisted Structure-Based Virtual Screening of Compounds That Target 15PGDH

Background: Skeletal muscle (SM) plays a critical role in movement, metabolism, and organ protection, with its maintenance and regeneration relying on muscle satellite (stem) cells (MSCs). Prostaglandin E2 (PGE2) regulates MSCs, but PGE2 levels decline with aging due to increased catabolism by 15-hydroxyprostaglandin dehydrogenase (15PGDH), a negative regulator of muscle repair. Methods: This study aimed to employ artificial intelligence and machine learning (ML)-assisted, structure-based screening approaches to identify novel 15PGDH inhibitors. Supervised models (support vector machine, random forest, and XGBoost were trained on curated bioactivity data (IC50 values) from the ChEMBL database and used to virtually screen the Maybridge compound library (~51,000 compounds). Results: The area under the curve (AUC) values of the developed models SVM, RF, and XGBoost were 0.96, 0.99, and 1.00, respectively. Promising inhibitors were further validated using structure-based virtual screening (docking), molecular dynamics simulations (200 ns), and MM-PBSA/GBSA analyses. The top five inhibitors (PD00616, HTS11491, HTS02629, AW00889, and HTS11190) were identified as active (ML analysis) and potential 15PGDH inhibitors based on their subsequent binding affinities, involvement of catalytic residues (Ser138, Tyr151, and Lys155), and complex stability. Additionally, these inhibitors were found to follow the drug-likeness criteria. Conclusions: These findings offer valuable insights for the development of novel therapeutics targeting 15PGDH to combat muscle degeneration and related pathologies, including aging and sarcopenia.