Higgins-Chen · Nature aging 2022 · Computational methodology and validation study · n=?

A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking.

Cited 527 times in the scientific literature.

Level 4 - case-series / case-control

Computational method development and measurement reliability analysis (graded by design analogy)

PubMed 36277076 · doi:10.1038/s43587-022-00248-2 · record verified 2026-08-29

What was done

The authors quantified technical variation across replicates for six prominent DNA methylation-based epigenetic clocks. To reduce this noise, they developed a computational approach that extracts principal components from CpG-level methylation data as inputs to retrain biological age prediction models. They evaluated the performance of these principal-component clocks on replicate agreement, association detection, intervention responsiveness, and longitudinal tracking in vivo and in vitro.

What was found

Standard epigenetic clocks exhibited replicate deviations of up to 9 years due to technical noise. Retrained principal-component versions of the six clocks reduced replicate discrepancies to within 1.5 years for most replicates. The authors also reported improved detection of clock associations, intervention effects, and more reliable longitudinal trajectories, without requiring technical replicates or prior CpG-reliability data during training.

Why it matters

Substantial measurement noise in traditional epigenetic clocks can mask true biological changes and treatment signals in clinical trials or longitudinal studies. This computational modification substantially improves measurement reliability and can be retroactively or prospectively applied to existing and new epigenetic biomarkers.

Limits

The abstract does not report the number of samples, datasets, or replicates used, nor does it specify the demographic characteristics or the exact interventions tested. Quantitative validation metrics beyond the 9-year vs. 1.5-year replicate deviation comparison are omitted in the abstract.

Cited by