Machine learning-driven identification of pan-PI3K inhibitors: A hybrid virtual screening approach combining naïve...

The phosphatidylinositol 3-kinase (PI3K) pathway is a vital intracellular signaling cascade that plays a key role in cancer cell survival, angiogenesis, and metastasis. Consequently, PI3K has been a primary target for therapeutic inhibition in the treatment of various malignancies. Accumulating studies indicate that targeting multiple isoforms of PI3K may enhance antitumor activity. The advancement of pan-PI3K inhibitors for cancer treatment offers a promising research and development opportunity. This study introduces a machine learning-based virtual screening (VS) method that combines a Naïve Bayesian classification model employing molecular fingerprints and descriptors, alongside a pharmacophore model and consensus scoring-based molecular docking, to discover new pan-PI3K inhibitors. The VS method was validated for its strong predictive accuracy by successfully identifying the commercially available pan-PI3K inhibitor copanlisib. The hybrid VS strategy was applied to the SPECS database, leading to the discovery of several promising PI3K inhibitor compounds. The machine learning-based VS approach is expected to offer meaningful insights and a practical framework for discovering novel PI3K inhibitors.