Predictive QSAR modeling of compounds inhibiting cytochrome P450 3A4-mediated metabolism of testosterone and 7-benzyloxy-4-trifluoromethylcoumarin

Enzymes in the cytochrome P450 family are responsible for much of the first-pass metabolism of xenobiotic compounds. Within this family, the hepatic 3A4 isoform (CYP3A4) is responsible for the first-pass metabolism of over half of the drug compounds currently on the market. This substrate promiscuity increases the risk of dangerous drug-drug interactions (DDIs), in which a drug compound inhibits the metabolism of other compounds by CYP3A4, leading to drug inactivity or the accumulation of the non-metabolized drug in the body. These risks have led to numerous quantitative structure-activity relationship (QSAR) and SAR studies of CYP3A4 inhibitors to determine the structural characteristics common to inhibitor compounds. Evidence of multiple binding pockets necessitates the use of a variety of probe substrates, resulting in different. From the published literature and patents, we collected compounds with inhibition data against CYP3A4, using either 7-benzyloxy-4-trifluoromethylcoumarin (BFC) or testosterone (TST) as the probe substrate and measuring inhibition as –log10(IC50) (pIC50). We then developed QSAR models using two descriptor selection methods (random forest and genetic algorithm-k nearest neighbors (GA-kNN) and two descriptor sets (MOE and Dragon). The resulting eight models were validated via five-fold cross validation and external validation. While the cross-validation results are good for all models, most models had low external predictivity. By analyzing the models with the best external predictivity (those using Dragon descriptors and GA-kNN descriptor selection), we found several atom-type and P-VSA-like descriptors that showed a sizable difference in importance between the models from the BFC and TST data. These descriptors reflect studies from prior QSAR studies on characteristics of CYP3A4 inhibitors. Results from this study could be used to account for differences in in vitro inhibition screens using multiple probe substrates.M.S.Includes bibliographical referencesby Christopher Mayer-Baco