DNA methylation quality control decisions, removed beta values, and their influence on epigenetic clocks

Measures of biological aging based on DNA methylation (epigenetic clocks) are commonly used in research across the biological, health, and social sciences. Many decisions are made during the quality control (QC) of the data on which such clocks are based, generating removed beta values that must be accommodated in construction of the clocks.

We apply a range of detection p-value and bead count thresholds during QC of a DNA methylation dataset and characterize the removed beta values. We then test different methods of imputing removed beta values and their impact on epigenetic clocks.

We find that both detection p-value and bead count thresholds remove beta values that differ significantly in their distribution from the values retained post-QC. Epigenetic clocks calculated from datasets using various imputation methods (mean, median, KNN, and methyLImp) do not appear to have consistent patterns of bias. The only exception is imputation of 0 for removed beta values (akin to leaving out those CpGs on a per-sample basis when calculating clocks), which shows stronger proportional bias of clock values for all clocks.

We recommend imputing removed beta values rather than leaving those CpGs out of clock calculations on a per-sample basis.