Machine Learning and Molecular Modeling for Drug Repurposing Targeting Potential PI3Kα Inhibitors in Post-CoViD-19 Pulmonary Fibrosis.

Dysregulation of the phosphoinositide 3-kinase-alpha (PI3Kα) pathway is implicated in the development of post-CoViD-19 pulmonary fibrosis, highlighting the need for effective therapeutic agents. This study aimed to identify novel PI3Kα inhibitors by computationally repurposing FDA-approved drugs. We employed a hybrid approach that combines machine learning with molecular modeling. A random forest (RF) classification model was built and validated using a curated data set of 4,023 known PI3Kα inhibitors from the ChEMBL database, demonstrating robust predictive performance. The RF model was applied to screen the subset of FDA-approved drugs available in the DrugBank database to identify potential candidates. The top-ranked compounds were subsequently evaluated through molecular docking, extensive 200 ns molecular dynamics simulations (MDS), and binding free energy calculations using the molecular mechanics/Poisson-Boltzmann surface area (MM/PBSA) method. Our virtual screening identified five promising drugs, with simeprevir and ceritinib demonstrating the most favorable free energy binding affinities (ΔG bind = -33.0 ± 3.2 kcal/mol and -25.2 ± 2.4 kcal/mol, respectively) and stable interactions within the enzyme's kinase domain. These findings highlight simeprevir and ceritinib as strong candidates for PI3Kα inhibition, warranting further experimental investigation for their potential use in treating post-CoViD-19 fibrotic conditions.