Machine learning-based identification of targeted metabolomic biomarkers for early diagnosis and fibrosis-stage discrimination in metabolic dysfunction-associated steatotic liver disease.

Background: Metabolically dysfunction-associated steatotic liver disease (MASLD) is the most prevalent liver disease worldwide and is increasing in parallel with metabolic syndrome and obesity. In this exploratory, cross-sectional study, we analysed metabolic changes in the blood and urine of patients with early-stage MASLD (fibrosis grades 0, 1, and 2) to identify metabolites associated with disease presence and the prevalent fibrosis stage and to develop machine learning models for case discrimination and fibrosis-stage stratification. Methods: Fifty-one metabolites, including17 urinary organic acids, 14 blood amino acids, 19 blood acylcarnitines/free carnitine, and glucose, were quantified in 232 participants (100 controls and 132 patients with MASLD: 68 F0, 34 F1, and 30 F2) using gas chromatography-mass spectrometry (GC/MS) and tandem mass spectrometry (MS/MS). MASLD-associated metabolites were visualised using volcano plots, cluster heatmaps, and a metabolic network diagram. Three machine learning approaches, namely, orthogonal partial least squares discriminant analysis (OPLS-DA), random forest (RF), and support vector machines (SVM), were implemented within a strictly leakage-free pipeline (feature selection and preprocessing performed within training folds only), with performance evaluated on independent test sets and validated by permutation testing and repeated cross-validation. Results: A case-discrimination panel Conclusion: Machine learning applied to targeted blood and urinary metabolomics identified candidate metabolite signatures associated with early-stage MASLD and fibrosis stage. These findings are exploratory and hypothesis-generating; prospective, externally validated studies with metabolically matched comparators are required before clinical application.