DNA methylation-based health predictors: advances, applications, and perspectives

Abstract

DNA methylation (DNAm) has emerged as a powerful and dynamic biomarker for predicting health outcomes, biological aging, and disease risk. Unlike static genetic variants, DNAm is dynamic and influenced by environmental, lifestyle, and pathological factors, making it highly suitable for applications in personalized medicine. This review provides a comprehensive synthesis of recent advances in DNAm-based predictors, including epigenetic clocks, exposure biomarkers, disease risk models, and trait-specific estimators. We describe the diverse methodological frameworks underpinning these predictors, such as penalized regression, surrogate modeling and deep learning. We discuss their performance across various preprocessing strategies and study populations. Additionally, we highlight clinical and research applications, ethical considerations, and emerging challenges, such as issues of reproducibility, tissue specificity, population generalizability, and interpretability. Looking forward, we explore future directions emphasizing artificial intelligence, multiomics integration, and longitudinal modeling. By critically assessing current limitations and technological innovations, this review outlines a roadmap for advancing the development, validation, and responsible implementation of DNAm-based health predictors.