ABSTRACT Aging is a global issue that affects human health and increases disease risk. The traditional concept of the “age gap (AG),” defined as the difference between estimated biological age and an individual's chronological age, has been used for self‐monitoring the risk of age‐related diseases. However, the current AG does not account for the stratified aging patterns across different stages of chronological age, which may lead to biased or paradoxical interpretations of aging acceleration. To address these limitations, we propose Personalized‐context‐Aware Age Gap (PAAG), a robust metric to estimate aging acceleration, based on our new pre‐training model AOE‐Net (Age Order Enhanced Network). AOE‐Net employs age‐order enhanced contrastive learning on multi‐omics data from healthy populations to learn latent representations that accurately reconstruct aging trajectories by capturing biological deviation rather than technical deviation in omics data. We demonstrate that PAAG, generated via fine‐tuning AOE‐Net, significantly outperforms AG of conventional first‐ and second‐generation aging clocks in predicting clinical outcomes. This superior predictive power was validated across diverse age‐related diseases and phenotypes: pan‐cancer (overall survival), subclinical atherosclerosis (PESA score), and osteoporosis (bone mineral density). Crucially, PAAG serves as a context‐aware metric that may improve the clinical outcome prediction of existing aging clocks. Furthermore, interpretive analysis of PAAG's molecular drivers revealed a strong functional enrichment for immune‐response pathways, providing a shared mechanistic link between accelerated aging and disease. Collectively, PAAG could serve as a stable indicator of aging acceleration for clinically assessing age‐related diseases, and AOE‐Net provides an effective pre‐training model for aging study and PAAG evaluation.
Wang et al. (2026) studied this question.