Identifying Platelet Lipidomic Networks and Evaluating Machine-Learning Models to Identify Distinctive Features Between Chronic and Acute Coronary Syndrome

Platelet lipidomics offers a window into the thromboinflammatory responses to acute coronary syndromes (ACS) and chronic coronary syndromes (CCS), yet differences between the platelet lipidomes of these two coronary artery disease subtypes remain poorly characterized. In this pilot study, untargeted platelet lipidomics and miRNA transcriptomics were performed on platelets isolated from 19 ACS and 57 CCS patients undergoing coronary angiography, integrated with routine clinical and hematological traits in statistical network analyses and cross-validated machine learning models. Platelet lipidomics identified 81 lipid features significantly altered between ACS and CCS (FDR q < 0.05), predominantly involving phosphatidylcholines, lysophosphatidylcholines, and ceramides. Cer 18:2;O2/24:0 showed the strongest inverse association with ACS of any identifiable lipid (q = 0.0394, OR = 0.297, 95% CI [0.171, 0.515]), while increased Cer 18:0;O2/22:1 (q = 0.088, OR = 3.072, 95% CI [1.607, 5.878]) and Cer 18:0;O2/18:0 (q = 0.088, OR = 2.791, 95% CI [1.515, 5.145]) demonstrate nominally significant promotive effects on ACS. Differential correlation analysis further identified oxidized phospholipid species as connectivity hubs in ACS-specific lipid networks, a pattern not detectable by differential abundance approaches alone. Although untargeted lipidomics did not meaningfully outperform a routine clinical trait benchmark in ACS classification, phosphatidylcholine-trained models achieved modest gains in AUROC, and the prominent ceramide signal—platelet-specific and mechanistically distinct from plasma ceramide risk scores—supports a targeted validation strategy centered on a ceramide-enriched platelet lipid panel in prospective, adequately powered cohorts.