Integrating QSAR, molecular dynamics and MM-GBSA to re-rank natural product–derived DPP4 candidates from virtual screening

Dipeptidyl peptidase-4 (DPP4) is a validated target for type 2 diabetes, and virtual screening of natural product libraries is a common route to new chemotypes. We asked whether a top docking rank survives independent scrutiny. From COCONUT, 85,600 compounds were filtered to 32,008 and docked against DPP4; 873 outscored the crystallographic ligand N7F (−10.01 kcal/mol), with Lig_771 first at −12.5 kcal/mol. Across four 300 ns simulations (two replicas per complex, 1.2 μs) Lig_771 stayed bound but never reproduced the gliptin recognition motif: the control held its salt bridges to Glu205 and Glu206 in over 93% of frames, whereas Lig_771 formed no persistent hydrogen bond to these residues and sampled two distinct poses. MM-GBSA gave −15.6 kcal/mol for Lig_771 against −23.9 for the control. A QSAR model built on 4327 measured DPP4 inhibitors (scaffold-based external R2 = 0.415; random split 0.669) predicted a pIC50 of 6.59 for Lig_771, below linagliptin (8.29), sitagliptin (7.39) and vildagliptin (6.96). None of the three analyses corroborated the docking ranking. The study is computational and does not establish DPP4 inhibition. Lig_771 occupies largely unexplored chemical space (maximum Tanimoto similarity 0.316) and merits testing, but selecting hits on docking score alone would have been misleading here.