Devising reliable and accurate epigenetic predictors: choosing the optimal computational solution

Abstract

ARISING FROM A.T. Higgins-Chen et al. Nature Aging https://doi.org/10.1038/s43587-022-00248-2 (2022) [1] Illumina DNA methylation arrays are frequently used for the discovery of methylation signatures associated with aging and disease [2]. With the latest Infinium Methylation EPIC BeadChips v1.0 and v2.0 covering 850-900K CpG sites across the human genome, one of the major hurdles to overcome when training trait prediction models is the high dimensionality of the data, with the number of features (target CpGs) greatly exceeding the typical number of samples assessed. In addition, most large-scale DNA methylation-based studies do not include replicate measurements for a given sample, making it impossible to estimate the degree of measurement uncertainty. Hence, the robustness of the assay and reliability of the prediction models are critical to ensure potential clinical utility. Recently, Higgins-Chen et al [1] proposed that training models on principal components (PCs) calculated from the data rather than directly on a selection of original features, results in more reliable age predictions. We found that dimension reduction with PCA consistently improves the reliability of age prediction models but may compromise their predictive accuracy. Our model projections further suggest that the balance in the trade-off between accuracy and precision we observed, may tip over in favor of PC-based models with increasing training sample numbers.