VAM: A Multimodal Dynamical Foundation Model for Characterizing Human Aging Dynamics and Enabling Virtual Aging Perturbation

Delaying aging and preventing age-related diseases require a precise, dynamic understanding of human aging. While biological age is widely used to quantify aging status, most existing methods rely on static estimates and fail to capture aging’s nonlinear dynamics or support in silico perturbation. Here, we propose the Virtual Aging Model (VAM), a multimodal framework that unites foundation models with dynamical network biomarker theory. Validated across the UK Biobank, NSPT, and VPPG cohorts, VAM captures nonlinear aging dynamics while its derived representations enable the identification of aging-associated molecular patterns and support in silico perturbation to prioritize aging-modulating candidates. Furthermore, we show that model-derived indices are associated with higher risk of multiple chronic diseases, indicating potential utility as early indicators of systemic instability. Collectively, our study presents VAM as an integrative framework that unifies aging quantification, dynamic modeling, and hypothesis generation for intervention strategies, thereby providing a systems-level view of aging and laying the foundation for future integrative and translational research.