Integrative metabolomics and machine learning identify biomarkers of adolescent social anxiety disorder

Social anxiety disorder (SAD) is one of the most prevalent anxiety disorders in adolescents but remains underdiagnosed due to the lack of objective diagnostic tools. This study aimed to identify serum metabolomic alterations in adolescent SAD patients and to develop an interpretable diagnostic model.

In this cross-sectional study, serum samples were collected from 78 adolescents, including 42 drug-naive, first-episode SAD patients and 36 matched healthy controls. Untargeted metabolomic profiling was performed, and feature selection was conducted via least absolute shrinkage and selection operator regression, followed by logistic regression for model construction.

Among the 661 detected metabolites, 46 differed significantly between groups, mainly within amino acid and energy metabolism pathways. Five key metabolites, 2-hydroxybutanoic acid, L-alanine, L-asparagine, glutamine and beta-tocopherol, were selected for model construction. The diagnostic model achieved an area under the curve of 0.934 in the training set, but external validation is still lacking, and the findings should be interpreted as hypothesis-generating.

Adolescents with SAD exhibit distinct metabolic profiles, and a preliminary diagnostic model was developed. Exploratory microbiota-related observations suggested potential links between gut microbial activity, host metabolism, and anxiety phenotypes, but these findings remain preliminary and outside the scope of the present study. Overall, these findings provide hypothesis-generating support for further investigation of gut-metabolism-brain interactions and highlight the need for larger, externally validated studies to advance biomarker development.