AI-Driven Identification of Bioactive Compounds in Medicinal Plants for Diabetes Management: An Integrative Computational and Validation Study

Abstract

Diabetes mellitus is a major worldwide social health problem, and, thus, the continued discovery of therapeutically safe, effective and cost-effective agents is necessary. Medical plants have traditionally been a good source of bioactive compounds with antidiabetic properties; however, the traditional discovery systems are limited by high financial expenditure, time-consuming processes, and the frequent occurrence of compounds that have been described and characterized before. In the current study, an artificial intelligence (AI)-based paradigm was used to identify and rank bioactive molecules produced through medicinal flora to treat diabetes. An edited list of ten antidiabetic plant species and twenty plant chemical entities was built. Quantitative structure-activity relationship (QSAR) models based on machine learning were used to predict both α-glucosidase inhibitory activity and the likelihood of inhibiting the dipeptidyl peptidase-4 (DPP-4) inhibition. Molecular docking schemes were run using structure-based protocol which determined binding affinity against α- glucosidase followed by in-silico ADMET profiling. Compounds with high rankings in computation were confirmed by synthetic in-vitro tests. The results have shown that several constituents such as thymoquinone, ellagic acid, Gymnemic acids and quercetin exhibited strong predicted inhibitory activities, good docking energies and acceptable pharmacokinetic properties. This paper highlights the effectiveness of AI-directed screening at accelerating the identification of natural product-derived antidiabetic therapeutics and provides a reproducible framework that can be used in future translational studies.