SIRT2i_Predictor: A machine learning-based tool to facilitate the discovery of novel SIRT2 inhibitors

Abstract

Selective sirtuin 2 (SIRT2) inhibitors hold therapeutic promise for treatment of wide range of age-related diseases. Despite promising preclinical results, none of the SIRT2 inhibitors have reached clinical trials. In order to facilitate development of novel SIRT2 inhibitors, a machine learning based tool titled SIRT2i_Predictor was developed through this work. The main utility of SIRT2i_Predictor is to support virtual screening (VS) campaigns and facilitate the selection of candidates for in vitro and in vivo evaluation. Appealing web-based interface which allows visualization of structure-activity relationships makes SIRT2i_Predictor a valuable tool in the lead optimization projects as well. The tool was built on panel of high-quality machine learning regression-based and binary classification-based models for prediction of inhibitors potency, as well as multiclass classification-based models for predictions of inhibitors SIRT1-3 isoform selectivity. The regression and classification structure-activity relationship models were created for 1797 publicly available compounds by exploring combinations of 5 machine learning algorithms and 4 molecular representations. SIRT2i_Predictor was demonstrated to be able to screen around 200000 compounds in matters of minutes with comparable chemical space coverage to the structure-based VS. The tool was applied in screening of in-house database of compounds further corroborating the utility in prioritization of compounds for costly in vitro screening campaigns. The code of SIRT2i_Predictor is made available at https://github.com/echonemanja/SIRT2i_Predictor.