QSAR model of pancreatic lipase inhibition by phenolic acids and their derivatives based on machine learning and multi-descriptor strategy

Abstract

Inhibition of pancreatic lipase can be effective in controlling or treating obesity. Phenolic acids have good pancreatic lipase inhibitory effects and are widely found in plants. The identification of phenolic acids and their derivatives that can efficiently inhibit pancreatic lipase has attracted widespread attention. Therefore, it is crucial to screen natural pancreatic lipase inhibitors efficiently by establishing quantitative structure-activity relationships (QSAR) models. In this study, machine learning was used to perform multiple computational analyses of 28 phenolic acids and their derivatives with pancreatic lipase inhibitory activity, and multiple QSAR models were established. Molecular parameters were analyzed, IC50 was predicted, and molecular docking and molecular dynamics simulations analysis were done for the screened natural compounds. The results showed that the model with the highest level of accuracy is the 2D-3D mix model with a numerator parameter value of 8 (R2fitting of 0.9537 and MAE of 0.2230), and MATS6c is a key descriptor. Further screened 5 compounds that have not yet been reported and have the potential to inhibit lipase activity (IC50pre ≤ 16.69 μM), these all bound to key residues (Tyr114 and Ser152) of pancreatic lipase, of which 3–4,1,5-carboxylic acid bound the most tightly (affinity of −9.8 kJ/mol). And the compounds-pancreatic lipase complexes (2-3-2-enoate) had good stability. The three compounds with high bioavailability have the potential to become novel pancreatic lipase inhibitors. This study provides a novel approach for the prediction of phenolic acids and their derivatives capable of inhibiting pancreatic lipase.