Uncertainty quantification in epigenetic clocks via conformalized quantile regression.

Area of Science:

• Epigenetics
• Biostatistics
• Computational Biology

Background:

• DNA methylation (DNAm) is a key epigenetic modification influencing biological age.
• Epigenetic clocks, based on DNAm, estimate biological age but often lack uncertainty quantification.
• Existing clocks predict mean biological age but do not fully capture population heterogeneity or provide prediction intervals.

Purpose of the Study:

• To develop a novel pipeline for training epigenetic clocks that incorporates uncertainty quantification.
• To reveal population heterogeneity and construct accurate prediction intervals for biological age.
• To improve the understanding of biological aging beyond mean estimations.

Main Methods:

• Integration of high-dimensional quantile regression and conformal prediction for epigenetic clock training.
• Development of a general pipeline applicable to various DNAm datasets.
• Application to 728 blood samples across 11 DNAm datasets from children and adolescents.

Main Results:

• The proposed method generates adaptive prediction intervals with nominal coverage and accounts for individual variability.
• Quantile regression-based prediction intervals were narrower than those from conventional mean regression, indicating improved statistical efficiency.
• The intervals revealed synchronized patterns with age acceleration, uncovering cellular evolutionary heterogeneity in aging during childhood and adolescence.

Conclusions:

• Conformalized high-dimensional quantile regression offers a robust framework for valid prediction intervals and uncovering population heterogeneity in epigenetic aging.
• The methodology enhances statistical efficiency and provides deeper insights into biological age distribution beyond the mean.
• This approach has broad applicability for understanding epigenetic age across age groups and informing anti-aging interventions.