Discovery of novel DGAT1 inhibitors by combination of machine learning methods, pharmacophore model and 3D-QSAR model

Abstract

DGAT1 plays a crucial controlling role in triglyceride biosynthetic pathways, which makes it an attractive therapeutic target for obesity. Thus, development of DGAT1 inhibitors with novel chemical scaffolds is desired and important in the drug discovery. In this investigation, the multistep virtual screening methods, including machine learning methods and common feature pharmacophore model, were developed and used to identify novel DGAT1 inhibitors from BioDiversity database with 30,000 compounds. 531 compounds were predicted as DGAT1 inhibitors by combination of machine learning methods comprising of SVM, NB and RP models. Then, 12 agents were filtered from 531 compounds by using the common feature pharmacophore model. The 3D chemical structures of the 12 hits coordinated with surface charges and isosurface have been carefully analyzed by the established 3D-QSAR model. Finally, 8 compounds with desired properties were retained from the final hits and have been assigned to another research group to complete the follow-up compound synthesis and biologic evaluation.

Keywords: 3D-QSAR; DGAT1; Machine learning method; Pharmacophore model; Virtual screening.