Development of pharmacophore-based classification model for activators of constitutive androstane receptor

Abstract

Constitutive androstane receptor (CAR) is predominantly expressed in the liver and is important for regulating drug metabolism and transport. Despite its biological importance, there have been few attempts to develop in silico models to predict the activity of CAR modulated by chemical compounds. The number of in silico studies of CAR may be limited because of CAR's constitutive activity under normal conditions, which makes it difficult to elucidate the key structural features of the interaction between CAR and its ligands. In this study, to address these limitations, we introduced 3D pharmacophore-based descriptors with an integrated ligand and structure-based pharmacophore features, which represent the receptor-ligand interaction. Machine learning methods (support vector machine and artificial neural network) were applied to develop an in silico model with the descriptors containing significant information regarding the ligand binding positions. The best classification model built with a solvent accessibility volume-based filter and the support vector machine showed good predictabilities of 87%, and 85.4% for the training set and validation set, respectively. This demonstrates that our model can be used to accurately predict CAR activators and offers structural information regarding ligand/protein interactions.

Keywords: Classification; Constitutive androstane receptor; Machine learning; Pharmacophore-based descriptor; Support vector machine.