Computational discovery of ATP-competitive GSK3β inhibitors using database-driven virtual screening and deep learning.

Abstract

Glycogen synthase kinase 3 beta (GSK3β) is a pivotal serine/threonine kinase implicated in diverse pathological conditions, making it a compelling target for therapeutic intervention. In this study, we employed a structure-based drug discovery approach to identify novel ATP-competitive GSK3β inhibitors through a multi-tiered computational framework. Reported inhibitors from various repositories were systematically analysed to establish physicochemical and interaction-based filters, facilitating the rational curation of screening candidates. Toxicity assessment via Derek Nexus further refined the selection, yielding seven lead compounds with optimal docking scores, robust interaction profiles, and adherence to drug-likeness criteria. Molecular dynamics simulations over 300 ns validated the stability of protein-ligand complexes with root mean square deviation, radius of gyration, and binding free energy calculations, substantiating sustained interactions. Key residues, including Lys85, Asp133, and Val135, were identified as critical for ligand stabilisation, corroborating reported inhibitor-binding mechanisms. Additionally, a deep learning based prediction model, GSK3BPred, was developed to classify potential GSK3β inhibitors. The GSK3BPred model is publicly available at https://github.com/PGlab-NIPER/GSK3BPred.git . This integrative computational strategy offers valuable insights into the discovery of novel ATP-competitive GSK3β inhibitors and lays a foundation for future experimental validation and optimization.