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

MetabHub integrates UK Biobank plasma NMR metabolomics and longitudinal outcomes, reporting 2,950,764 significant associations and 1,886,224 mediation pathways across 57 incident diseases.

• Why it matters: It addresses the gap that existing resources separately cover metabolite–disease links, QTLs, or prediction, with limited integrated mediation and longitudinal risk stratification support.
• What they did: Built from 364,795 European participants with 251 NMR traits, 129 clinical measurements, 54 exposures, genome-wide variants, and mediation via two-step screening and natural effect models. It also trains metabolite-based incident-disease predictors (penalized logistic regression, XGBoost, LightGBM, multilayer perceptron) and evaluates risk scores with Cox regression and survival stratification over ~14 years.
• The result: The platform enables mechanistic discovery and metabolomics-informed precision prevention by combining mediation pathway identification with pre-trained prediction and ~14-year survival risk stratification.