High-throughput DeepPRM-Stellar proteomics coupled with machine learning enables precise quantification of...

Predicting the progression of asymptomatic large-artery atherosclerosis (LAA) to acute ischemic stroke (AIS) remains a significant challenge when relying solely on anatomical stenosis. To address this clinical gap, we integrated discovery-phase serum proteomics with machine-learning techniques to identify circulating biomarkers capable of predicting atherosclerotic progression. Utilizing a dual-cohort design (Cohort I: discovery stage, n = 43; Cohort II: validation stage, n = 39), we established a Serum Protein Candidate Biomarker Bank (SPCBB) encompassing 1484 proteins by harmonizing literature-derived evidence (1369 proteins) with 222 differentially expressed proteins (DEPs) identified through mass spectrometry analysis. Global proteomics revealed that LAA-associated proteins were enriched in cholesterol metabolism, whereas AIS was characterized by the activation of complement/coagulation cascades. We performed targeted validation of 171 peptides (corresponding to 156 proteins) using DeepPRM on the Stellar platform, thereby facilitating machine learning-based optimization of the biomarker panel. The XGBoost algorithm identified two diagnostic signatures: a three-protein panel (RNASE4, HBA1, ATF6B) that differentiates AIS from LAA, with an area under the curve (AUC) of 0.917 and specificity of 80.0%; and a six-protein panel (MRC1, HBA1, GUC2A, HBD, CLEC3B, FLNA) that distinguishes AIS/LAA from healthy controls, achieving an AUC of 0.971 and specificity of 86.0%. To further validate key candidates, we performed ELISA assays for GUCA2A and FLNA, which confirmed their significant elevation in patients with AIS and LAA (p < 0.01), consistent with the proteomics findings. Both internal and external validations confirmed robust performance across cohorts. These validated biomarker panels establish a proteomics-driven framework for serum-based, dynamic monitoring of LAA progression, thereby supporting clinical decision-making aimed at optimizing early stroke prevention in asymptomatic individuals.