Machine Learning-Driven Prediction of Coronary Artery Disease Risk Based on UK Biobank Plasma Proteomics

Area of Science:

• Cardiovascular Medicine
• Genomics
• Proteomics
• Biomarker Discovery

Background:

• Coronary artery disease (CAD) remains a leading cause of mortality globally.
• Conventional risk models for CAD exhibit limited predictive accuracy.
• There is a need for integrated approaches combining diverse data types for improved risk prediction.

Purpose of the Study:

• To develop and validate a unified model for enhanced coronary artery disease (CAD) risk prediction.
• To integrate conventional risk factors, polygenic risk scores, and large-scale proteomics data.
• To assess the incremental predictive value of proteomic data in CAD risk stratification.

Main Methods:

• Utilized UK Biobank data, including plasma proteomics and genetic risk data, excluding prevalent CAD cases.
• Trained CatBoost models incorporating conventional risk factors, polygenic risk scores, and a 202-protein proteomic risk score.
• Employed least absolute shrinkage and selection operator (LASSO) Cox regression for risk score derivation and Shapley Additive Explanations (SHAP) for feature selection, identifying a 9-protein panel.

Main Results:

• The proteomic risk score demonstrated a dose-dependent association with CAD risk across validation cohorts.
• Integration of polygenic and proteomic risk scores significantly improved CAD risk discrimination compared to conventional factors alone (AUC increased from 0.750 to 0.789 in internal validation).
• A compact 9-protein panel, including GDF15, MMP12, and ACE2, captured substantial proteomic predictive information.

Conclusions:

• Integrating conventional risk factors, polygenic risk scores, and proteomic data substantially enhances CAD risk prediction.
• Proteomics plays a crucial role in advancing precision cardiovascular medicine.
• The developed models and identified protein panels offer potential for simplified and more accurate cardiovascular risk stratification tools.