From virtual screening to bench: A dual-validation framework for drug repurposing against PI3K.

Virtual screening has emerged as one of the most impactful in silico approaches for the identification of novel drug candidates, substantially reducing the cost and time associated with high-throughput screening (HTS). Ongoing efforts focus on exploring large-scale libraries of drug-like molecules to identify candidates with favourable pharmacological properties. In this study, we propose an applicability domain-based virtual screening strategy that extends beyond conventional approaches by prioritising compounds with ADMET profiles comparable to marketed drugs. To further enhance predictive performance, we developed a QSAR model on PI3K ligands using Light Gradient Boosting Machine (LGBM), which achieved an R2 value of 0.799, thereby providing an additional layer of validation for compound selection. The phosphoinositide 3-kinase (PI3K) pathway, a critical regulator of cell growth, survival, metabolism, and proliferation, is frequently dysregulated in multiple cancers and other diseases. Repurposing existing drugs that modulate PI3K activity offers the potential to accelerate therapeutic development while mitigating the challenges of de novo drug discovery. To demonstrate the utility of our approach, we screened two compound libraries from Enamine-a hit-like locator library (>400,000 molecules) and a kinase-focused library (>64,000 molecules)-against the PI3K-α isoform. In addition, a set of 1367 FDA-approved drugs was screened to identify potential candidates for repurposing. From these extensive datasets, three small molecules from the Enamine libraries were identified with favourable drug-like properties and synthetic accessibility compared with existing PI3K-α inhibitors. Furthermore, one FDA-approved drug demonstrated potential PI3K-α inhibitory activity. Pharmacophore mapping provided additional validation of their drug-likeness. Importantly, wet-lab evaluation of the FDA-approved drug confirmed its inhibitory activity, thereby supporting the computational predictions. Overall, our integrated in silico and experimental framework highlights promising PI3K-α inhibitors, underscoring the potential of applicability domain-based virtual screening and QSAR modelling for both drug discovery and repurposing.