Unexpected metabolism can lead to the failure of many late-stage drug candidates or even the withdrawal of approved drugs. Therefore, during early research, it is important to predict the sites of metabolism and metabolites of potential drug-like molecules.

Historically, predictive models have targeted metabolism by human Cytochrome P450 (CYP) isoforms due to their well-documented importance in phase I metabolism. Here, we will present methods to predict isoform-specific metabolism for a broad range of enzymes involved in phase I and phase II metabolism, including aldehyde oxidases (AOs), flavin-containing monooxygenases (FMOs), sulfotransferases (SULTs) and UDP-glucuronosyltransferases (UGTs) alongside CYPs. These models are based on a consistent framework, combining mechanistic quantum-mechanical simulations with machine learning, and are rigorously validated with experimental data. The resulting models predict if a potential site of metabolism on a compound is likely to be metabolised by the specified enzyme.

We will demonstrate how these site-of-metabolism models can be combined with models that predict which enzymes and isoforms are likely to metabolise a compound. Applying models iteratively to a parent compound and its metabolites enables the prediction of metabolic pathways and the resulting metabolites observed in vivo. We validate these pathway predictions by comparison with experimentally observed metabolite profiles.

The figure shows the metabolic pathway predicted for Dextromethorphan using the heuristics. The blue background represents the parent compound, the green background represents the reported metabolites, and the white background represents the predicted metabolites that were not reported experimentally. The figure was generated with the StarDrop™ software platform.

Mario Oeren

Appearances