Model predicts age more accurately in tissue datasets
Trained on 93 multi-tissue datasets, the computational model made smaller age-prediction errors than comparison models, including in tests on eight datasets excluded from training.
Limit: Age-prediction accuracy is not a direct measure of biological aging
Edited by Boris Djordjevic · AgingResearch.News
A computational study used 93 datasets spanning multiple tissues to train a model that estimated age from DNA methylation, chemical marks on DNA. The model combined several neural networks—computer models that learn patterns—and adjusted how it used biologically labeled components for each sample.
In five-fold cross-validation, which repeatedly split the data into training and testing groups, the model had the lowest average and median sizes of prediction errors among matched comparison models. These measures ignored whether estimates were too high or too low. It also had the lowest average size of prediction errors across eight datasets kept out of training.
Why it matters
DNA methylation clocks are used to evaluate biological aging, disease risk and interventions. The approach offers a framework for exploring how biologically labeled pathways relate to age across tissues and disease contexts.
Caveats
These computational comparisons do not establish that more accurate age predictions better measure biological aging. The study is a preprint and has not been peer-reviewed.
The paper
University of Pennsylvania
bioRxiv, Oct 22, 2026, Open access, CC BY-NC-ND, Preprint, not peer-reviewed
