MetabHub: an integrative platform for plasma nuclear magnetic resonance metabolomics using mediation analysis, disease prediction, and risk stratification

Plasma nuclear magnetic resonance (NMR) metabolomics offers a scalable molecular readout linking genetic variation, environmental exposure, and clinical physiology to future disease risk. Existing resources typically address metabolite-disease associations, metabolite quantitative trait loci, or prediction separately, with limited support for integrated mediation and longitudinal risk stratification. MetabHub (https://www.metabhub.net) is a web-based resource built from individual-level UK Biobank data, comprising 364 795 European participants, 251 plasma NMR traits, 129 clinical measurements, 54 environmental exposures, genome-wide variants, and 57 incident diseases. It maps sex-combined and sex-specific associations across measurements, exposures, genotypes, metabolites, and diseases, yielding 2 950 764 significant associations. Using two-step screening and natural effect models, MetabHub identifies 1 886 224 significant clinical measurement, environmental factor, and genotype mediation pathways. It also provides metabolite-based prediction models for incident disease using penalized logistic regression, XGBoost, LightGBM, and multilayer perceptron, together with an online interface for applying pre-trained models to user-uploaded profiles. Model-derived metabolomic risk scores are further evaluated using Cox regression and survival stratification over ~14 years of follow-up. Global search, interactive visualization, downloadable tables, prediction, and stratification modules support mechanistic discovery and metabolomics-informed precision prevention.