Understanding the metabolism of small molecules is critical for the development of safe and effective drugs. Computational approaches can help in several ways, including predicting metabolically labile atoms in a molecule, known as sites of metabolism (SoMs), and predicting the chemical structures of a molecule’s metabolites. A common approach to metabolite structure prediction is to generate a set of possible metabolites based on reaction rules that describe known metabolic reactions. One challenge with this type of approach lies in determining which – out of a potentially large number of generated metabolites – would be more likely to be observed experimentally. We have developed an approach to address this challenge by integrating machine learning-based SoM prediction into the prediction of metabolite structures. The basis for the machine learning component of this approach is FAME 3, a collection of machine learning models for SoM prediction that have demonstrated high prediction accuracy for both phase I and phase II metabolism.<br/><br/>This talk presents GLORYx, a tool to predict and rank the chemical structures of metabolites that could be formed by phase I and phase II metabolism. In extending beyond CYP-mediated metabolism, which is the domain of our previous approach called GLORY, we needed to address the additional complexity of ranking the predicted metabolites from different enzyme families. To this end, GLORYx employs machine learning models from FAME 3 along with new reaction type-specific SoM models in order to achieve better ranking of the generated metabolites. Validation on a high quality, manually curated test set has shown that incorporating SoM prediction models in this way can help prioritize the metabolites that are generated from a set of reaction rules.

GLORYx: Using machine learning to predict metabolites for phase I and phase II metabolism

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

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