A novel serum metabolite classifier for identifying Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD) integrating metabolomics and machine learning.

BACKGROUND: Metabolic dysfunction-associated steatotic liver disease (MASLD) presents a growing global health burden, while reliable non-invasive biomarkers for identifying affected individuals remain limited. This study aimed to characterize serum metabolomic signatures associated with steatosis and fibrosis and to develop a clinically applicable metabolite-based classifier. METHODS: Untargeted serum metabolomics was performed in a FibroScan-characterized discovery cohort (n = 35) to identify candidate metabolic features associated with hepatic steatosis and fibrosis burden. To provide biological context for the observed metabolic alterations, 16 S rDNA sequencing was subsequently conducted in paired fecal samples from a subset of participants (n = 27). Differential metabolites were then subjected to LASSO regression for feature selection and used to construct a random forest diagnostic model in a validation cohort comprising healthy controls (n = 19) and ultrasound-confirmed MASLD patients (n = 52). RESULTS: Metabolomic profiling revealed distinct metabolic patterns across different degrees of steatosis and fibrosis. A total of 55 and 46 metabolites were identified as differentially abundant in relation to steatosis and fibrosis burden, respectively. Microbiome analysis indicated alterations in gut microbial composition, and integrative correlation analysis suggested several potential microbe-metabolite associations, including two metabolite-genus pairs showing relatively strong correlations. LASSO regression selected a panel of ten metabolites as the most informative diagnostic features. Using these metabolites as input variables, a random forest classifier was constructed and achieved an area under the receiver operating characteristic curve (AUROC) of 0.87. Incorporation of four clinical variables (BMI, ALT, triglycerides, and HDL cholesterol) further improved model performance, yielding an AUROC of 0.94. CONCLUSION: This study characterizes systemic metabolic alterations associated with MASLD and presents a non-invasive diagnostic model integrating serum metabolites with clinical indicators. Together, these findings highlight the potential utility of metabolomics-assisted approaches for identifying MASLD and for improving biological understanding of disease-associated metabolic changes.