Integrating Imaging-Derived Clinical Endotypes with Plasma Proteomics and External Polygenic Risk Scores Enhances...

Area of Science:

• Cardiology
• Genetics
• Medical Imaging

Background:

• Coronary microvascular disease (CMVD) contributes significantly to ischemic heart disease but is underdiagnosed.
• Development of risk prediction models for CMVD is hindered by limited large-scale genome-wide association studies (GWAS).
• There is substantial genetic overlap between CMVD and coronary artery disease (CAD), suggesting shared genetic risk factors.

Purpose of the Study:

• To develop polygenic risk score (PRS) models for CMVD using external CAD GWAS data.
• To integrate PRS, plasma proteomics, and perfusion PET imaging for CMVD risk prediction using machine and deep learning.
• To establish an unsupervised, imaging-based endotyping framework for CMVD to identify distinct patient subgroups.

Main Methods:

• Developed CMVD PRS models by selecting variants from CMVD GWAS and applying weights from CAD GWAS.
• Integrated plasma proteomics, perfusion PET imaging measures, and PRS into machine and deep learning models.
• Created an unsupervised endotyping framework using perfusion PET-derived myocardial blood flow data.

Main Results:

• Integrated multimodal data (genetics, proteomics, imaging) improved CMVD risk prediction models.
• The novel imaging-based endotyping framework revealed distinct CMVD patient subgroups.
• This stratification significantly enhanced classification performance (AUROCs 0.65-0.73 per class) compared to binary classifiers and clinical models.

Conclusions:

• Imaging-based endotyping combined with multimodal data offers a promising approach for precise CMVD diagnosis.
• This stratification strategy captures CMVD heterogeneity, enabling more personalized diagnostic and treatment strategies.
• This study establishes a framework for multimodal modeling in complex cardiovascular diseases, integrating genetic, proteomic, and imaging data.