TidyMass2: advancing LC-MS untargeted metabolomics through metabolite origin inference and metabolic feature-based functional module analysis

Untargeted metabolomics provides a direct window into biochemical activities but faces critical challenges in determining metabolite origins and interpreting unannotated metabolic features. Here, we present TidyMass2, an enhanced computational framework for Liquid Chromatography-Mass Spectrometry (LC-MS) untargeted metabolomics that addresses these limitations. TidyMass2 introduces three major innovations compared to its predecessor, TidyMass: (1) a comprehensive metabolite origin inference capability that traces metabolites to human, microbial, dietary, pharmaceutical, and environmental sources through integration of 11 metabolite databases containing 532,488 metabolites with source information; (2) a metabolic feature-based functional module analysis approach that bypasses the annotation bottleneck by leveraging metabolic network topology to extract biological insights from unannotated metabolic features; and (3) a graphical interface that makes advanced metabolomics analyses accessible to researchers without programming expertise. Applied to longitudinal urine metabolomics data from human pregnancy, TidyMass2 identified diverse metabolites originating from human, microbiome, and environment, and uncovered 27 dysregulated metabolic modules. It increased the proportion of biologically interpretable metabolic features from 5.8% to 58.8%, revealing coordinated changes in steroid hormone biosynthesis, carbohydrate metabolism, and amino acid processing. By expanding biological interpretation beyond MS2 spectra-based annotated metabolites, TidyMass2 enables more comprehensive metabolic phenotyping while upholding open-source principles of reproducibility, traceability, and transparency. Untargeted metabolomics faces challenges in metabolite origin inference and metabolite annotation. Here, the authors present TidyMass2, a user-friendly computational framework with metabolite origin inference and feature-based functional module analysis to enhance biological interpretation.