DNA-methyaltion-based age predictor

The study presents a software tool for predicting chronological age based on DNA methylation patterns. Key points:

1. The tool uses a large training dataset of 13,661 DNA methylation samples from blood and saliva.

2. The age predictor's accuracy improves as the training sample size increases, potentially leading to a near-perfect predictor with sufficient data.

3. The association between the age acceleration residual (AAR, the difference between predicted and actual age) and mortality weakens as the predictor's accuracy improves.

4. Based on the Elastic Net algorithm, the best predictor shows no significant association between AAR and mortality in the Lothian Birth Cohorts of 1921 and 1936.

5. Predictors trained on smaller sample sizes are more susceptible to confounding by cellular compositions than those trained on larger sample sizes.

6. The predictor performs comparably in non-blood tissues to a multi-tissue-based predictor.