Integrative Molecular Scaffold Generation of PI3K/mTOR Inhibitors Using DeepSARM and Ligand-Based Virtual Screening Approaches.

Designing novel core scaffolds for small molecules is an important yet challenging task in medicinal chemistry, particularly during the lead-generation stage and in situations that require alternative chemotypes and/or chemical novelty. In this study, we applied an artificial intelligence-based structure generation using the DeepSARM methodology to create novel scaffold series and to augment medicinal chemistry capabilities. Samotolisib, a PI3K/mTOR dual inhibitor featuring a tricyclic scaffold, was selected as a model ligand. Following DeepSARM-guided structure generation and ligand-based virtual screening, several candidate compounds with simplified scaffolds were identified. Notably, one top-ranked virtual hit 6 was selected and further optimized into a simplified tool compound 6A for experimental validation. Subsequent synthesis and in vitro evaluation led to the identification of a validated hit bearing a novel nontricyclic scaffold.