Identification of novel potential hypoxia-inducible factor-1α inhibitors through machine learning and computational simulations.

Introduction: Hypoxia-inducible factor-1α (HIF-1α) has become a significant therapeutic target for breast cancer and other cancers by regulating the expression of downstream genes such as erythropoietin, thereby improving cell survival in hypoxic conditions. Methods: We jointly applied a multistage screening system encompassing machine learning, molecular docking, and molecular dynamics simulations to conduct virtual screening of the "Traditional Chinese Medicine Monomer Library" for potential HIF-1α inhibitors. The virtual screening was conducted in three sequential stages, applying the following selection criteria sequentially: an activity prediction score greater than or equal to 0.8, a stronger binding affinity, and an MM-PBSA binding free energy lower than the reference compound. Results and Discussion: We retrieved 361 compounds with HIF-1α inhibitory activity data from the ChEMBL database for the construction and evaluation of machine learning models. Among the six constructed models, the random forest model based on RDKit molecular descriptor with the optimal comprehensive performance was employed for virtual screening. Ultimately, four compounds were selected for binding mode analyses and 100 ns molecular dynamics simulations. The results showed that the compounds Arnidiol and Epifriedelanol exhibit the most stable interactions with the HIF-1α protein, which can serve as potential HIF-1α inhibitors for future investigations.