Development of chemoinformatics method for predicting drug metabolites

Abstract

It is indispensable to investigate ADMET(absorption, distribution, metabolism, elimination, and toxicity) properties in the drug discovery process. Generally, examinations of metabolism take a long time in ADMET test process. Therefore, the method for predicting a metabolite of a drug candidate is strongly needed. In this study, we developed the method for predicting metabolites using chemoinformatics methods. In order to predict metabolites, we have to reveal mainly two selectivities: isoform specifity, which is the selectivity whether a molecule is metabolized by enzymes, and regioselectivity, which is the selectivity of the position where metabolic reaction preferentially occurred in a molecule by enzymes. We built both isoform specificity models and regioselectivity models by using several machine learning methods such as support vector machine and random forest. By applying isoform specificity models for a drug candidate, we can predict enzymes which metabolite it. By applying regioselectivity models for it, we can predict site of metabolism of the compound. Through this process, we can predict metabolites of drug cadidates.