Interpretable machine learning and molecular simulations identify natural pancreatic lipase inhibitors and hydrophobic hotspot residues.

Pancreatic lipase (PL) is a validated peripheral target for limiting dietary fat absorption, yet structurally diverse natural inhibitors remain scarce. We assembled a PL inhibitor dataset from public resources and trained a random-forest classifier using PubChem fingerprints test AUC = 0.9134 to prioritize a natural product library. After drug-likeness and toxicity filtering, docking, and experimental validation, three hits were identified: Licochalcone C IC = 62.05 ± 1.77 μM, Neoglycyrol IC = 95.21 ± 0.57 μM, and Notopterol IC = 104.27 ± 3.19 μM. Interaction fingerprint and molecular dynamics analyses showed that binding was dominated by hydrophobic interactions, with Val260/Ala261 acting as key residues across the three PL-ligand complexes. Dissociation free-energy profiles from steered molecular dynamics and umbrella sampling were consistent with the potency ranking. Collectively, this data-driven pipeline identified new natural PL inhibitors and provided residue-level insights for further optimization.