Key result
Machine learning with lipidomics outperforms clinical models in predicting functional stenosis, achieving an AUC of ~0.81.
Why the study?
Invasive angiography and quantitative flow ratio primarily reflect focal anatomical hemodynamics and fail to capture systemic metabolic vulnerability, motivating the development of a noninvasive predictive framework using integrated lipidomics and machine learning for functional stenosis and multivessel disease.
Does an integrated lipidomics and machine learning framework improve the noninvasive prediction of functional coronary stenosis and multivessel disease compared to conventional clinical models in patients with acute coronary syndrome?
Cohort (n=441)
Does an integrated lipidomics and machine learning framework improve the noninvasive prediction of functional coronary stenosis and multivessel disease compared to conventional clinical models in patients with acute coronary syndrome?
Integrating plasma lipidomics with machine learning (specifically Support Vector Machine) outperforms conventional clinical models in noninvasively predicting functional coronary stenosis and multivessel disease in patients with acute coronary syndrome.
Background While invasive angiography and quantitative flow ratio (QFR) are gold standards for diagnosing coronary functional ischemia, they primarily reflect focal anatomical hemodynamics and fail to capture the systemic metabolic vulnerability of the patient. We aimed to develop and rigorously validate a noninvasive predictive framework using integrated lipidomics and machine learning (ML) for functional stenosis and multivessel disease. Methods High-throughput plasma lipidomic profiling (LC-MS/MS) was performed on a cohort of 441 patients with confirmed acute coronary syndrome. We integrated high-dimensional lipid data with clinical features to construct and optimize ML models (including Support Vector Machine [SVM], Random Forest, and XGBoost). To evaluate real-world robustness against data drift, the optimal models underwent rigorous prospective temporal validation based on a chronological enrollment cutoff. Results The SVM model demonstrated superior performance, achieving cross-validated area under the curve (AUC) values of 0.81 for functional stenosis and 0.79 for multivessel disease, significantly outperforming conventional clinical models. Crucially, prospective temporal validation confirmed the models' robust generalizability (AUC 0.66 with high positive predictive value). Explainable AI (SHAP analysis) successfully pinpointed specific atherogenic lipid drivers: ether-linked phosphatidylinositol (ePI 38:4) emerged as the dominant predictor for general stenosis, reflecting intense oxidative stress, whereas specific sphingomyelins (SM 38:5) and phosphatidylserines (PS 36:1) characterized the multivessel phenotype, biologically mirroring macrophage foam cell formation and the expansion of the apoptotic necrotic core. Conclusion This ML-driven lipidomic framework enables robust, noninvasive systemic risk stratification for functional ischemia and multivessel disease. By uncovering specific atherogenic lipid signatures, it not only optimizes pre-catheterization triage but also identifies potential novel lipid targets for future precision cardiovascular therapeutics.Figure 2.Temporal Validation of ML Models.Figure 1.Abstract
No takes yet. Share an insight, caveat, or question.
Bao et al. (2026) conducted a cohort in Acute coronary syndrome (n=441). Integrated plasma lipidomics and machine learning models (SVM, Random Forest, XGBoost) vs. Conventional clinical models was evaluated on Prediction of functional coronary stenosis and multivessel disease (AUC 0.81 (functional stenosis); AUC 0.79 (multivessel disease)). Integrated lipidomics and SVM machine learning predicted functional stenosis (AUC 0.81) and multivessel disease (AUC 0.79) in ACS patients, outperforming conventional clinical models.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: