This study investigates a DNA methylation predictor designed to estimate chronological age using a panel of 209 cytosine-phosphate-guanine (CpG) sites. The research is grounded in the analysis of whole-blood samples, utilizing both development and external cohorts to validate the predictor’s accuracy.
The primary methodology involved the application of a five-nearest-neighbor reconstruction (KNN5) combined with Ridge regression to enhance the prediction of chronological age. The study included 524 development samples and 132 held-out samples from the GSE40279 dataset, along with an external cohort from GSE73103 consisting of 355 samples.
Results indicated that the intact-profile prediction achieved a mean absolute error of 4.24 years and a root-mean-square error (RMSE) of 5.38 years, with a strong correlation (Pearson r = 0.934) between predicted and actual ages. The study also examined the predictor’s performance under conditions of empirical and random missingness, revealing that KNN5 significantly reduced RMSE by 0.365 years compared to mean replacement methods, particularly under higher missingness burdens.
Further analyses demonstrated that the reconstruction advantage persisted even when sample-level and CpG-level failure counts were randomized. External predictions maintained a correlation with chronological age (r = 0.835), despite some calibration bias that varied with age.
Overall, the findings support the utility of a stability-selected CpG panel for predicting chronological age, even when faced with incomplete data profiles. This research contributes to the understanding of DNA methylation as a biomarker for aging and highlights the importance of methodological robustness in age prediction models.
What this study does not show
This study does not establish the clinical utility of the DNA methylation predictor for health outcomes, as it focuses solely on chronological age estimation without assessing health-related implications.
Evidence scorecard
- Analytic validity: established
- Clinical validity: partial
- Clinical utility: not established
The evidence is based on primary research in human samples.