Machine learning-aided serum N-glycomic signatures reveal potential biomarkers for acute ischemic stroke and minocycline treatment response.

The absence of molecular-level tools for diagnosing and monitoring acute ischemic stroke (AIS) underscores the need for novel biomarkers. Serum N-glycosylation represents a promising but remains underexplored source of such markers in AIS. Using hydrophilic interaction liquid chromatography with fluorescence detection (HILIC-FLD), serum N-glycome profiling from healthy controls (HC), AIS patients who had not received prior treatment (AIS) and administered minocycline (A_M) were conducted. A computational pipeline integrating multiple machine-learning (ML) models and feature-selection algorithms was employed to decode the glycomic data. AIS was associated with a distinct serum N-glycome signature characterized by increased branching and hypersialylation, which was partially reversed after the minocycline treatment. An optimized random forest model (Glyco_RFF) based on six glycome features (Glyco_F: GP14, GP32, GP33, GP34, S3, S4) discriminated both AIS from HC, and AIS from A_M with high accuracy (AUC values > 0.9). Longitudinal changes in two features (GP34, S4) correlated significantly with clinical improvement (NIHSS score reduction), suggesting their potential as dynamic biomarkers of treatment response. This study identifies a dynamic serum N-glycome signature in AIS and its modulation by minocycline. The Glyco_RFF model provides a novel, non-invasive tool for diagnosis and therapeutic monitoring.