Predicting xenobiotic metabolism: a computational approach mining LC-MS/MS data with SIRIUS and BioTransformer

Drug biotransformation and bioactivation play a pivotal role in drug discovery, driving the development of analytical methods to investigate xenobiotic metabolism. However, the identification of drug metabolism products (i.e., drug metabolites) remains challenging. Drug metabolites are difficult to predict, and they are often missed without prior knowledge of the drug's metabolic fate. Untargeted approaches overcome this requirement, but demand strategies for metabolite identification. In this study, we developed a computational workflow, using high resolution LC-MS/MS metabolomics data, integrating BioTransformer and SIRIUS for the prediction and putative identification of drug metabolite structures. We challenged our workflow to the analysis of human metabolites from 6 well-known drugs, administered to primary human hepatocytes and human liver microsomes: amitriptyline (10 µM), carbamazepine (12.5 µM), cyclophosphamide (20 µM), fipronil (20 µM), phenytoin (50 µM), and verapamil (6 µM). Of the drugs' metabolites, 62-100% were found using this computational approach. Furthermore, 4 new metabolite structures (1 amitriptyline metabolite and 3 verapamil metabolites) were proposed using de novo predictions in SIRIUS. This strategy proved efficient in accelerating the study of drug metabolism, potentially avoiding tedious manual metabolite identification. In sum, we demonstrate that this computational workflow holds potential in automating metabolite identification, expanding metabolite coverage, and elucidating metabolites of newly developed drugs.