Reverse Biotransformation-Guided Annotation of Untargeted MS/MS Features: A Computational Framework for Candidate Enzyme and Gene Hypothesis Generation

Untargeted LC–MS/MS experiments detect thousands of metabolic features, yet most remain unannotated due to limited spectral and structural database coverage. To address this limitation, we present a modular workflow that identifies closest structural analogs from MS/MS-derived features, applies rule-based reverse biotransformation to infer plausible precursor reactions, and maps these reactions to candidate enzymes and genes. Confidence-aware parameter selection and rule scoring are incorporated to balance annotation coverage with biological plausibility. Evaluation on reference data sets using systematic sensitivity analyses justified analog retrieval and biotransformation confidence thresholds. Pipeline-derived gene sets consistently recapitulated pathways reported in prior metabolite-gene association studies and exhibited stronger pathway enrichment than baseline associations. Application to independent experimental metabolite lists produced biologically coherent pathway enrichments across heterogeneous data sets. For a lesser-characterized metabolite, inferred genes were highly enriched in glutathione metabolism and oxidative stress pathways (adjusted p < 2 × 10 –4 ). This confidence-aware integration of spectral annotation and reverse biotransformation provides a reproducible and interpretable framework for generating candidate enzyme and gene hypotheses from poorly annotated MS/MS features, enhancing the biological interpretation of metabolomics-driven discovery.