Machine learning-based QSAR and structure-based virtual screening guided discovery of novel mIDH1 inhibitors from natural products

Mutations in isocitrate dehydrogenase 1 (IDH1) have been widely observed in various tumors, such as gliomas and acute myeloid leukemia, and therefore has become one of the current research focal points. Therefore, it is crucial to find inhibitors that could target mIDH1, which may provide more effective treatment options for patients with related tumors. In present study, combines machine learning-based QSAR models and structure-based virtual screening to screen a series of potential IDH1 inhibitors from the Coconut databases. The QSAR model predictions indicate that the hit compounds have high binding affinity to the target protein, and its pIC₅₀ value was found to be considerably larger than that of AGI-5198. The RMSD and Rg analysis demonstrated that all of the ligand–protein complexes exhibited a stable state throughout the simulation period. Furthermore, the binding free energy decomposition and per-residue contribution of the IDH1^R132H-inhibitor complex revealed key fragments of the inhibitor interacting with residues ALA-111, PRO-118, ARG-119, LE-128, ILE-130, ITRP-267, VAL-281, and TYR-285 in the binding site of IDH1^R132H. This investigation indicates that CNP0047068, CNP0029964, and CNP0025598 have the potential to be targeted inhibitors of IDH1^R132H mutants through further optimization, providing new insights for discovering novel lead scaffolds in this domain.