Integrative machine learning and molecular simulation strategies for BCR-ABL inhibition in chronic myeloid leukemia

Chronic myeloid leukemia (CML) is predominantly driven by the oncogenic fusion protein BCR-ABL1, with tyrosine kinase inhibitors (TKIs) representing the primary therapeutic approach. Nevertheless, the emergence of the T315I mutation in the ABL1 kinase domain markedly reduces the effectiveness of TKIs, posing a major therapeutic challenge. In this study, we present a robust
in silico
workflow aimed at discovering potent and selective inhibitors against the T315I-mutated ABL1 kinase. The strategy combines machine learning-assisted quantitative structure-activity relationship (QSAR) modeling with molecular docking, followed by extensive molecular dynamics (MD) simulations performed under both physiological (310 K) and stress-induced (500 K) temperature conditions to evaluate stability and binding dynamics. A total of 2,727 phytochemicals from the PhytoHub database were mapped to PubChem IDs and screened based on drug-likeness, binding affinity, docking scores, hydrogen bonding, and predicted IC₅₀ values. Among them, 15 compounds exhibited superior predicted activity (IC₅₀ < 31.65 nM) compared to the reference inhibitor, rebastinib. Notably, compound 5281238 (flavoxanthin) emerged as a top candidate, with a predicted IC₅₀ of 31.46 nM. MD simulations revealed its remarkable stability (RMSD = 0.15 nm over 100 ns), and interaction analysis confirmed key hydrogen bonding with HIS138. Free energy landscape (FEL) profiling and molecular mechanics general born surface area (MM/GBSA) analysis (ΔG_TOTAL = -61.91 kcal/mol) further supported its favorable binding profile. This study identifies Flavoxanthin as a promising lead for combating T315I-associated resistance in CML and advancing targeted therapy development.