Recent preclinical findings have identified selective inhibitors of sirtuin 2 (SIRT2) as potential therapeutic agents for treating age-related diseases, but none have advanced to clinical trials. The growing adoption of machine learning (ML) techniques in drug discovery has demonstrated their transformative potential, yet there remains a lack of large-scale, robust ML models for identifying novel SIRT2 inhibitors. To fill this gap, we developed SIRT2i_Predictor, a machine-learning-based tool designed to assist in virtual screening (VS), lead optimization, and the selection of SIRT2 inhibitors for experimental validation. The tool integrates a series of high-performance ML models, both for regression and classification, to predict the potency of inhibitors and their selectivity across SIRT1-3 isoforms. These models were trained on an extensive dataset comprising 1797 compounds using state-of-the-art ML algorithms. A comparison with traditional structure-based VS protocols revealed that the tool not only covers a comparable chemical space but also offers significant improvements in processing speed. The tool was successfully applied to screen an in-house compound database, confirming its utility in prioritizing candidates for costly in vitro testing. With a user-friendly web interface, SIRT2i_Predictor is accessible to the broader research community, and its source code is freely available online.
