FusionAge framework for multimodal machine learning-based aging clocks uncovers cardiorespiratory fitness as a major driver of aging and inflammatory drivers of aging in response to spaceflight

Traditional epigenetic aging clocks are limited because they do not incorporate clinical information and functional tests, and rely on DNA samples and methylation profiling infrastructure which are not easily accessible. To address these limitations, we built a new framework, FusionAge, with which we trained 26 aging clocks using interpretable nonlinear models, including deep neural networks (DNNs). Our results show that multimodal clocks built with DNNs significantly outperform clocks derived from single modalities or traditional linear models. FusionAge-derived biological age is more strongly associated with incident disease and mortality compared to chronological age in UK Biobank individuals. We validated these findings in the National Health and Nutrition Examination Survey, confirming that cardiorespiratory fitness is a major, consistent driver of biological age. Finally, we applied FusionAge to demonstrate its utility in detecting biological age changes in astronauts following spaceflight. Together, we demonstrate a powerful, portable framework for assessing biological age that captures the complex, multifactorial nature of human aging.