Prediction of cytochrome P450 2B6-substrate interactions using pharmacophore ensemble/support vector machine (PhE/SVM) approach.

An in silico model for predicting human cytochrome P450 2B6-substrate interactions was generated based on a novel scheme, which was initially devised to predict the hERG liability (reported in Leong, M. K., Chem. Res. Toxicol., 2007, 20, 217.) using pharmacophore ensemble/support vector machine to take into account the protein conformational flexibility while interacting with structurally diverse substrates. This is of critical importance yet never being addressed by any analogue-based molecular modeling studies before. Thirty-seven molecules were chosen from the literature and scrutinized for structural integrity and data consistency, of which 26 were treated as the training set to generate models, which were subject to validation by the other 11 molecules as the test set. The predicted pK(m) values by the final PhE/SVM model were in good agreement with observed values. In addition, this in silico model produced an r(2) of 0.84 and a 10-fold cross-validation q(2) of 0.66 for the training set and an r(2) of 0.87 for the test set, asserting the fact that this PhE/SVM model is an accurate model to predict the human P450 2B6-substrates interactions and can be used as a robust prediction tool to facilitate drug discovery.