Probabilistic prediction of the human CYP3A4 and CYP2D6 metabolism sites.

This article briefly introduces the results of in silico prediction of the most probable metabolism sites for the human cytochrome P450 3A4 and 2D6 isoforms. Ligand-based QSAR models have been developed using a novel GALAS modeling approach, and provide probabilities of being a target of CYP3A4 or CYP2D6 for any atom in a molecule. The GALAS-model development methodology allows evaluation of the reliability of predictions in the form of estimated prediction Reliability Indices (RIs). For all the models considered in this study, the number of misclassifications and inconclusive results was reduced significantly when only predictions of high quality (RI>0.5) were taken into account, demonstrating that RI reflects accuracy of prediction. The applicability domain of regioselectivity models is shown to be easily expandable to cover compound classes of interest to the user. The results obtained so far show promising perspectives for the utilization of the GALAS modeling in the analysis of regioselectivity for other important biotransformation enzymes--a work currently in progress.