Prediction of Human Cytochrome P450 Inhibition Using Support Vector Machines

Cytochrome P450s (CYPs) play a major role in the metabolism of drugs utilized in human health care. Inhibition of these enzymes by a drug may result in unwanted drug-drug interactions when two or more drugs are coadministered. Therefore, CYP inhibition should be investigated as early as possible in lead discovery and lead optimization. In silico approaches are highly desirable to assess large data sets or virtual compounds. Here we present the application of support vector machines (SVMs) to predict the potency of structurally diverse drug-like molecules to inhibit the human cytochromes P450 3A4 (CYP3A4) and 2D6 (CYP2D6). Different descriptor sets were used to cover various aspects of molecular properties, including physico-chemical properties derived from the 2D structure, the interactions of the molecule with its environment, and properties derived from quantum-mechanical calculations. Support vector classifiers were trained to distinguish between strong, medium, and weak inhibitors. For both isoenzymes, independent test set compounds were correctly re-classified with an accuracy of approximately 70%. The data sets were also used to generate support vector regression models. The best models were able to predict the log IC 5 0 values of the test set compounds with a squared correlation coefficient of R 2 = 0.67 (CYP3A4, corresponding RMSE of 0.36 log units) and R 2 =0.66 (CYP2D6, corresponding RMSE of 0.44 log units). Our results show that SVMs are a very powerful tool to predict CYP inhibition liability from calculated physico-chemical properties without invoking any information about the active site of the enzyme. The models can, for instance, be utilized to flag problematic compounds in an early step or to guide further synthesis efforts in a later stage of a research project.