Identifying farnesoid X receptor agonists by naïve Bayesian and recursive partitioning approaches

Abstract

The farnesoid X receptor (FXR), a ligand-modulated transcription factor, is a multiple functional hepatic cell protector. Therefore, FXR agonists present as promising dyslipidemia and anti-diabetes agents. To identify novel FXR agonists, models were created from 144 known FXR agonists by naïve Bayesian (NB) and recursive partitioning (RP) approaches. The predictive and reliable models were selected with Matthews correlation coefficient (MCC) criterion (>0.900 with 117 testing compounds). The top 4 models were validated with the external data (282 compounds having cell-free activities and 500 decoys). Two optimal FXR agonist models (one from the NB method and the other from the RP method) were obtained from the top models by further validation. A virtual screening campaign was conducted against our in-house compound library with the optimal models and produced 15 virtual hits, which were further confirmed with cell-based luciferase assays. Finally, we discovered two new FXR agonists. Molecular docking studies indicated that the two new FXR agonists have similar binding modes to the known FXR agonists. This work demonstrated that a machine learning approach with combined NB and RP methods was able to identify novel FXR agonists and that the approach could be applied in other lead identification processes.