Liquid biopsy, cfDNA, and machine learning: an integrative approach to cardiovascular diseases

Abstract

Cardiovascular diseases (CVDs) and their complications remain the leading cause of mortality worldwide. However, obtaining diseased cardiovascular tissue is often invasive, costly, and clinically impractical, highlighting the need for noninvasive approaches that can capture dynamic pathological information. Cell-free DNA (cfDNA)-based liquid biopsy has emerged as a promising strategy for the diagnosis, monitoring, and risk assessment of CVDs. As cfDNA is released from injured, dying, or stressed cells, it carries multiple layers of biological information, including concentration changes, tissue-of-origin signatures, fragmentation patterns, nucleosome footprints, mitochondrial DNA signals, and epigenetic modifications. These signals can reflect global cellular injury, identify tissue origin, and indicate inflammatory or mitochondrial stress. Current cardiovascular evidence is unevenly distributed: donor-derived cfDNA has shown the strongest clinical relevance in heart transplant rejection surveillance; cfDNA concentration, cardiomyocyte-specific methylation, and fragmentomic features show promise in myocardial infarction, heart failure, and selected aortic diseases, whereas applications in hypertension, valvular disease, congenital heart disease, and cardiomyopathy remain exploratory. Among cfDNA features, methylation is especially informative because it provides stable tissue- and disease-associated signatures; however, methylation datasets are typically high-dimensional, sparse, and affected by low disease-derived cfDNA fractions. Machine learning may help address these challenges by selecting informative features, integrating methylation with fragmentomic and nucleosomal signals, and developing models for diagnosis, prognosis, tissue-of-origin inference, and longitudinal monitoring. In this review, we summarize the biological characteristics of cfDNA, evaluate its current applications in major CVDs, highlight how machine learning may improve cfDNA interpretation, and discuss key barriers to clinical translation.