Development and validation of QSAR-based machine learning models for screening of DPP-IV inhibitors

DPP-IV inhibitors have been largely used for the treatment of type 2 Diabetes worldwide. Although it has been shown their effectiveness in reducing glycated haemoglobin, they do not present any weight loss properties, possibly due to not inhibiting solely the GLP-1 substrate. Therefore, in this work we developed QSAR-based machine learning models for the screening of GLP-1-selective DPP-IV inhibitors. Initially, 6,306 compounds were selected from ChEMBL database. After rigorous data curation and balancing of the dataset, six QSAR models were developed and validated. The best models presented correct classification rate and sensitivity above 0.80 were used for the screening of DPV-IV inhibitors from the commercial dataset provided by ChemBridge Corporation. The results suggest that the developed MACCS and Avalon models might be effective for the screening of DPP-IV inhibitors, but not the Atom Pair, Morgan, Sirms and Dragon models. The selected virtual hits will be purchased and experimentally validated.