The ability to understand and predict the metabolic stability of xenobiotics is of key importance to a host of different industries. This contribution will start with a brief overview of the scope and limitations of current in silico models for the prediction of (i) interactions of xenobiotics with metabolizing enzymes, (ii) sites of metabolism (i.e. atom positions where metabolic reactions are initiated) and (iii) likely metabolites. We will then report on our current efforts in exploring various machine learning strategies and data sources to develop more accurate models with a broader applicability domain. In particular, we will report on the further development of FAst MEtabolizer (FAME), a random forest-based predictor of sites of metabolism in small molecules that obtains high robustness by using a small set of circular atom descriptors. We will also present a new metabolite structure predictor that makes use of predictions of metabolically labile atom positions by FAME to significantly reduce the false-positive prediction rates, which pose a big challenge to current in silico approaches. In addition, we will report on the development of a new model for predicting the interaction of xenobiotics with metabolizing enzymes trained on a large curated dataset compiled from several public and proprietary sources.

Machine learning models for the prediction of xeniobiotic metabolism

256th National Meeting (2018)

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