MetabolEx: Platform for Small-Molecule Metabolites Generation and Metabolic Pathway Prediction

Accurate metabolite prediction is essential for identifying safety liabilities and optimizing pharmacokinetic properties during early drug discovery. However, predicting metabolic fate remains a major bottleneck due to the risk of unexpected toxicity and drug-drug interactions. We present MetabolEx, a platform that integrates rule-based generation with machine learning prioritization to predict Phase I and II metabolism. MetabolEx utilizes 194 manually curated reactions in SMIRKS notation and machine learning classifiers trained on 41,041 substrate-metabolite pairs from multiple sources. This hybrid approach enumerates chemically plausible metabolites, while applying probabilistic filtering to manage the combinatorial complexity of recursive prediction. External validation on 219 substrates demonstrated the platform's performance across three recursive prediction levels. At L1, MetabolEx achieved 0.84 recall for the primary metabolites. At L2, the tool reached double the precision of state-of-the-art methods with a significant reduction in false positives (FP) while maintaining a substantial recall of 0.57. At L3, MetabolEx achieved 5-fold higher precision and a 25-fold reduction in FP compared to exhaustive enumeration, while maintaining a recall of 0.35. These results demonstrate that probabilistic filtering converts high-volume metabolic output into a focused and manageable set of predictions. MetabolEx captures complex transformations across eight reaction classes, including heterocycle metabolism and glutathione conjugation, and serves as a comprehensive resource for metabolic and toxicological profiling. The platform features interactive network visualization and automated toxicity annotation and is freely available at https://metabolex.exscalate.eu.