<p>Ischemic stroke (IS) is a leading cause of mortality and long-term disability worldwide, with ultra-early diagnosis critical for effective reperfusion therapy but challenged by nonspecific symptoms and limited sensitivity of imaging in the hyperacute phase. Plasma metabolomic alterations occur rapidly after ischemia onset, offering potential for accessible biomarkers and deeper insights into underlying pathogenic mechanisms. We therefore aimed to explore plasma metabolic alterations and their associated regulatory networks in IS. We conducted a multicenter case–control study involving 493 IS patients and 493 controls individually matched for age, sex, and body mass index (BMI). Untargeted LC–MS-based metabolomic profiling was performed on plasma samples to quantify and identify differentially abundant metabolites. Machine learning models, including LASSO and XGBoost, were used to select diagnostic biomarkers and construct a classification model. Model performance was evaluated using ten-fold cross-validation, an internal test set, and two independent external validation cohorts. Core regulatory genes were identified by integrating metabolite-related genes from HMDB with transcriptomic data from the GEO dataset. Functional enrichment, single-cell RNA sequencing, and drug-gene interaction analyses were further employed. We identified 319 endogenous differential metabolites, among which 18 core metabolites were selected to build an XGBoost diagnostic model. L-Methionine (Log2FC = 4.97), Purine (Log2FC = − 3.12), and Threonic acid (Log2FC = − 1.74) were the most influential contributors. The metabolite-based model showed strong internal discrimination under ten-fold cross-validation and internal testing; however, external validation was heterogeneous, with substantially reduced discrimination in the Ningbo cohort and preserved performance in the Suzhou cohort. This pattern suggests center-dependent variability and potential optimism in internally estimated performance, warranting cautious interpretation. Furthermore, ten core regulatory genes (e.g., MAOA, MSRB2, ACSM2A) were identified and implicated in neurotransmitter metabolism, oxidative stress, and fatty acid activation. Single-cell analysis revealed cell-type-specific expression patterns of the detectable core genes in endothelial cells and microglia. Drug prediction highlighted several repurposable compounds, including levodopa and tryptamine, with predicted binding affinity to target proteins. This multi-omics study characterizes a plasma metabolic signature associated with IS while highlighting cohort-dependent variability in external validation. Although internally derived diagnostic performance was high, heterogeneous external validation underscores the importance of cautious interpretation and prospective real-world validation. Overall, these findings should be regarded as hypothesis-generating and require confirmation in larger, consecutively recruited real-world populations.</p>