A machine learning approach to metabolomics identifies putative biomarker candidates and dysregulated pathways for distinguishing gout from asymptomatic hyperuricemia in the Zhuang population.

introductionGout typically develops from hyperuricemia (HUA), but the metabolic alterations driving this transition remain poorly understood, limiting our understanding of disease pathogenesis.

objectivesTo identify stage-specific putative biomarker candidates and to characterize dysregulated metabolic pathways distinguishing gout from HUA.

methodsWe conducted a targeted metabolomics assay on the baseline plasma samples from a Zhuang minority cohort using LC-MS/MS. The analyzed sample set comprised 38 HUA patients, 47 gout patients, and 52 healthy controls. Sex-stratified differential metabolite analysis was performed across all participants, as well as in female and male subgroups. Pathway enrichment analysis was carried out using the KEGG database. Machine learning approaches, including the Boruta algorithm and support vector machine (SVM), were employed for putative biomarker discovery and model evaluation in male participants.

resultsAmong all participants, 24 metabolites reached nominal significance (P < 0.05), but only uric acid remained significant after FDR correction. In sex-stratified analyses, no metabolite survived FDR correction in females, whereas in males, seven metabolites (flavone, glutamine, L-2-aminoadipic acid, L-pipecolic acid, N1-methyl-2-pyridone-5-carboxamide, phenyllactic acid, and uric acid) showed significant differences among healthy controls, HUA patients, and gout patients (FDR < 0.1). These metabolites were primarily involved in nitrogen metabolism, arginine biosynthesis, D-amino acid metabolism, nicotinate and nicotinamide metabolism, and purine metabolism. Machine learning identified four metabolites (N1-methyl-2-pyridone-5-carboxamide, flavone, glutamine, and phenyllactic acid) that distinguished gout from healthy controls, with AUCs of 0.902 and 0.800 in the training and validation sets, respectively. A second model (L-pipecolic acid, glutamine, phenyllactic acid, and flavone) discriminated gout from HUA, achieving AUCs of 0.850 and 1.000. Sensitivity analyses excluding obese or hypertriglyceridemic participants confirmed the robust performance of both models.

conclusionsThis study suggests sex-specific metabolic alterations in gout and provides robust machine learning-based models for male participants. The identified metabolite signatures appear to extend purine metabolism to involve amino acid and energy metabolic pathways. These findings provide a basis for mechanism-targeted strategies in HUA management. External validation remains essential.