Task-Aligned Foundation Models for Linking Imaging Features with DNA Methylation in Aging Research

Biological aging shapes infection risk, functional decline, and clinical outcomes. DNA methylation (DNAm) provides a complementary molecular readout of these processes by capturing host responses to infection, quantifying biological age via epigenetic clocks, and revealing conserved aging trajectories, while routine histopathology and emerging spatial imaging encode morphological features associated with these molecular and aging states. In this work, we develop a task-aligned methodological framework that unifies three analytical strands through an aging-centered lens: (1) formalizing image→ methylome tasks that infer CpG values, module scores, or methylation-defined phenotypes from histopathology or radiology with orthogonal molecular validation; (2) defining methylation-and imagingderived aging endpoints-including chronological age, DNAm age, biological-age acceleration, stratified performance, uncertainty quantification, and lifespan-wide calibration; and (3) outlining foundation-model strategies for multimodal integration, in which transformer-based architectures align molecular and morphological representations for reliable clinical deployment. Building on these components, we propose practical conventions for tasks, datasets, and evaluation metrics, including age-band reporting, uncertaintyaware calibration, and subgroup fairness across age, sex, and ancestry. Taken together, this framework provides a standardized methodological basis for developing trustworthy multimodal epigenetic clocks and identifies near-term opportunities for diagnostic stratification, disease staging, and longitudinal monitoring in older adults within both research and translational healthcare settings.