Deep-learning SASP score predicts mortality and disease
Trained on UK Biobank proteomic data, a deep-learning senescence score independently predicted mortality risk and shifted during an 18-month exercise trial.
Aging Cell
In human data from the UK Biobank Pharma Proteomics Project, researchers developed a composite score to track systemic cellular senescence burden using a semi-supervised deep learning framework. The team curated blood proteins from the senescence-associated secretory phenotype and analyzed them using a Guided autoencoder with Transformer model. The resulting score served as a strong, independent predictor of mortality risk and incident chronic medical conditions, including dementia, chronic obstructive pulmonary disease, myocardial infarction, and stroke. In an independent randomized clinical trial cohort, multimodal exercise significantly changed the trajectory of the score over 18 months.
Why it matters
Circulating senescence-associated proteins may offer a practical blood-based proxy for systemic cellular senescence burden. The findings suggest the score could help evaluate geroscience interventions and track risk for multiple chronic conditions.
Caveats
Circulating secretory proteins are a proxy for burden rather than an observation of senescent cells in tissues. Additionally, the abstract does not report the sample sizes or participant characteristics for the cohorts.
- Deep learning
- Exercise
- Cellular senescence
- Senescence-associated secretory phenotype (SASP)
- SASP Score
- Humans
The paper
Show 6 more authors
Julie Loebach Wetherell, Laura Haynes, Perla El Ahmad, Richard H. Fortinsky, George A. Kuchel, Trevor Harris,University of Connecticut
Aging Cell · 26 Sep 2026 · CC BY