Serum metabolic characteristics and potential biomarkers of first-episode schizophrenia: An exploratory machine...

Metabolomics studies in first-episode schizophrenia (FES) remain limited and heterogeneous. This study aimed to characterize the serum metabolomic profile of FES, identify metabolic biomarkers, and develop machine learning-based predictive models. In this exploratory study, serum metabolomic profiles were compared among healthy controls (HCs; n = 10), untreated FES patients (n = 10), and the same patients after second-generation antipsychotics (SGAs) treatment. Candidate biomarkers were evaluated using six machine learning algorithms. Bayesian network analysis and weighted gene co-expression network analysis (WGCNA) were further applied to explore the network relationships. Cross-dataset integrative analysis was additionally performed using an external public dataset comprising patients with schizophrenia (n = 30) and HCs (n = 30). Disease-associated candidate biomarkers included PG(16:0/18:1), PC(20:4/15:0), PC(22:5/15:0), PG(18:1/18:1), and PE(16:0/16:0), whereas treatment-associated biomarkers included PG(16:0/18:1), PC(20:4/15:0), PC(22:5/15:0), cholesterol sulfate, DG(20:4/22:4), and Phe-Thr-Asn-Lys. All selected candidate biomarkers achieved Area Under the Curve (AUC) values above 0.9, with performance varying across algorithms. In the Bayesian networks, disease served as a direct parent node of the Montreal Cognitive Assessment (MoCA) score and all biomarkers, whereas treatment shared direct edges with the Positive and Negative Syndrome Scale (PANSS) score and all biomarkers except cholesterol sulfate, which was indirectly connected via Phe-Thr-Asn-Lys. WGCNA identified metabolite modules significantly associated with disease status (r = -0.88, p < 0.01) and treatment (r = 0.75, p < 0.01). Cross-dataset analysis identified phenylpyruvic acid, SM(d18:1/22:1(13Z)), and 2-octenoylcarnitine as shared differential metabolites with modest-to-moderate discrimination. These exploratory findings reveal potential serum metabolic signatures of FES and provide a basis for the development of objective biomarkers for diagnosis and SGAs treatment monitoring, warranting validation in larger independent cohorts.