Cytochrome P450 (CYP) accounts for approximately 90% the metabolism of top prescribed medicines. Nine CYP isoforms--1A2, 2A6, 2B6, 2C19, 2C8, 2C9, 2D6, 2E1, and 3A4--metabolize the majority of orally administered drugs in adults while CYP 3A5 and 3A7 are important in neonates and young children. Understanding the combined metabolic activity of these 11 isozymes is critical to predict changes in pharmacokinetics and pharmacodynamics profiles in individual patients. <i>In vitro</i> experiments with human liver microsomes and <i>in vivo</i> animal models are commonly used to estimate the metabolism of a drug candidate in human. Unfortunately, these experiments are both expensive to comprehensively test the metabolism of vast numbers of candidates during early stages drug development and insufficient to use in personalized medicine. We train a deep neural network model on a literature-derived data set of 9764 compounds to predict specific metabolism by the aforementioned 11 CYP isozymes. Our model has major improvements over published computational metabolism models: given a test molecule, it can predict the sites (atoms/bonds) and the types of biochemical transformations, and the isozymes that contribute to the metabolism of that molecule. The model differentiates between metabolized and non-metabolite sites with cross-validated average-site accuracy of 97.4% and predicts metabolizing isozyme with average-molecule accuracy of 96.2%. The model developed in this study will be available at http://swami.wustl.edu/xenosite.

Deep learning isozyme-specific P450 metabolism

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.