AI-assisted molecular design of quinolone alkaloid analogues as pancreatic lipase inhibitor candidates

Pancreatic lipase (PL) is a key enzyme in dietary fat absorption and a validated target for preventing obesity. However, the existing PL inhibitors, such as orlistat, have side effects. We developed a molecular design workflow that integrates machine-learning-based generation with iterative docking simulations, focusing on quinolone-alkaloid-like scaffolds. The docking score of the best candidate decreased from -8.0 to -13.5 kcal/mol over 60 optimization cycles. The top-scoring structure contained a polycyclic aromatic core linked via conjugated chains, with docking poses indicating π–π stacking with Tyr114 and Phe215, consistent with a pocket-blocking mechanism that impeded substrate entry. Additionally, we observed a polar approach near Ser152. The generator produced chemotypes that resembled known inhibitors despite starting from quinolone alkaloid scaffolds. Although we focused on docking-based optimization, these findings demonstrate the potential of AI-assisted molecular design for identifying novel PL inhibitors. Further validation is required, including systematic structure-activity analyses and experimental testing.