MetabolEx combines curated metabolic reaction rules with machine-learning prioritization to make metabolite prediction more focused, interpretable, and actionable in early drug discovery.

Predicting metabolism is essential to anticipate clearance, exposure, drug–drug interactions, and safety liabilities before experimental testing. MetabolEx addresses this by combining curated SMIRKS reaction rules with machine-learning classifiers trained on validated substrate–metabolite relationships. Instead of exhaustively listing all chemically possible metabolites, it prioritizes the biologically plausible ones, reducing noise in recursive predictions while preserving mechanistic interpretability. The platform supports Phase I and Phase II metabolism prediction, interactive pathway visualization, and toxicity/reactivity annotation for safety-oriented profiling.

Highlights

·       From enumeration to prioritization: MetabolEx does not simply generate all possible metabolites; it ranks them according to their predicted biological plausibility, making the output more useful for decision-making.

·       Broad metabolic coverage: The platform captures major Phase I and Phase II biotransformations, including oxidation, dealkylation, hydrolysis, reduction, glucuronidation, sulfonation, methylation, and glutathione conjugation.

·       Performance and safety-oriented insight: MetabolEx showed strong ability to identify primary metabolites, reaching a recall of 0.84. More importantly, when metabolism prediction becomes recursively complex, the platform keeps the output interpretable: it achieved 5-fold higher precision and a 25-fold reduction in false positives compared with state-of-the-art tools, while also helping highlight reactive intermediates and toxicity-related metabolic liabilities that could otherwise emerge only later in development.

MetabolEx makes metabolite prediction more practical for early drug discovery. Its main value is not only predictive accuracy, but the ability to reduce the noise generated by recursive metabolism models. By combining metabolic pathway prediction, probabilistic prioritization, and toxicity annotation, MetabolEx supports earlier ADME and safety de-risking of small molecules.

The current version focuses on Phase I and Phase II hepatic metabolism. It does not yet cover transporter-mediated Phase III processes, microbiome-driven metabolism, or rare structural transformations. Another key future direction is site-of-metabolism prediction, which would help distinguish where a reaction is most likely to occur when multiple reactive sites are present in the same molecule.

MetabolEx provides a transparent and validated framework to predict the metabolic fate of small molecules, combining curated chemical knowledge with machine-learning prioritization. The platform is freely available online and the full paper describes its architecture, validation strategy, benchmarking results, and case studies in detail.

Discover more and try the platform at https://metabolex.exscalate.eu/metabolite. For further details, read the full paper here https://pubs.acs.org/doi/10.1021/acsmeasuresciau.6c00019.