Abstract
Cytochrome P450 3A4 (CYP3A4) is an important member of the CYP family and responsible for metabolizing a broad range of drugs. Potential drug-drug interactions (DDIs) caused by CYP3A4 inhibitors could lead to increasing risk of side-effects/toxicity or decreasing effectiveness. The evaluation of CYP3A4 inhibitory activity is time-consuming, labor-intensive, and costly, and it is necessary to establish virtual screening models for predicting CYP3A4 inhibitors. In this study, 4 classifier algorithms, including support vector machine (SVM), naive Bayesian (NB), recursive partitioning (RP), and K-nearest neighbor (KNN), were applied to discriminate CYP3A4 inhibitors from the non-inhibitors. Correlation analysis and stepwise linear regression methods were used for descriptor selection and optimization. The performance of classifiers was measured by 5-fold cross-validation, Y-scrambling and test set validation. Finally, the optimal NB model with Matthews correlation coefficients of 0.894 for the test set was developed to screen FDA-approved drugs and natural products database. As a result, 90 compounds from FDA-approved drug databases were predicted as inhibitors, and 46% of them were identified as known CYP3A4 inhibitors. 6 natural products were selected for further bioactivity assay and molecular docking. 2 of them with good docking score also exerted significant CYP3A4 inhibitory activities with IC50 values of 0.052 and 1.120 μM, respectively. This study proved the feasibility of a new method for predicting CYP3A4 inhibitory activity and preventing the occurrence of DDIs at early stage in drug development.