Computational antidiabetic assessment of Salvia splendens L. polyphenols: SMOTE, ADME, ProTox, docking, and molecular dynamic studies

Abstract

This study utilizes artificial intelligence and machine learning to enhance drug discovery, focusing on the antidiabetic effects of Salvia splendens leaf extract among the global epidemic of diabetes mellitus. Employing the SMOTE oversampling strategy confirmed that the generated dataset mirrored the activity pattern of the original data. An ADMET analysis of twelve compounds indicated that most complied with Lipinski's rule of five, demonstrating favorable oral bioavailability and safety profiles, except for two compounds, luteolin7- O -(4″,6″-di- O-α- L-rhamno-pyranosyl)- β -D-glucopyranoside and apigenin-7- O-β -D-rutinoside, which exhibited low solubility. Molecular docking studies on α -glucosidase and protein tyrosine phosphatase 1B revealed that compound 4 had the highest binding energy, surpassing that of the standard drug rosiglitazone. Molecular dynamic simulation studies indicated greater stability of docked α -glucosidase compared to tyrosine phosphatase after docking with the promising compounds. Overall, the findings highlight the potential of phenolic compounds from S. splendens as candidates for Type 2 diabetes management. • The study uses AI and machine learning to enhance drug discovery for diabetes treatment. • SMOTE oversampling produced a dataset that accurately reflects the original activity patterns. • Most evaluated compounds met Lipinski's rule of five, indicating good oral bioavailability, while two showed low solubility. • Molecular docking revealed that compound 4 surpassed the standard drug rosiglitazone in binding energy to α-glucosidase. • Dynamic simulations indicated that α-glucosidase maintained greater stability than protein tyrosine phosphatase after docking with the promising compounds.