Adapting Epigenetic Clocks for Cell-Free DNA High-Throughput Sequencing Data

Abstract Epigenetic clocks trained on methylation arrays generalize poorly to high-throughput sequencing (HTS) of cell-free DNA (cfDNA). Using paired array and HTS replicates, we systematically identified requirements to bridge this platform gap and developed a standardized, model-agnostic adaptation framework. Optimal performance requires maintaining at least 10× mean target depth, utilizing L2-weighted regularization, implementing targeted beta-value imputation and transfer learning. A combined framework using these strategies significantly enhanced legacy clock performance across independent aging and disease cohorts, enabling robust, minimally invasive biological age assessment without compromising biological interpretability. Competing Interest Statement W.H., X.Z., J.W., Y.G., Z.Y., S.J., B.H., D.T., V.T., C.T., X.F., C.O. and G.Z. are affiliated with Regenerative Bio, a global biotech company powered by AI, providing organ-specific aging assessment models and longevity-focused health solutions. Regenerative Bio was the study sponsor. The other authors declare no competing interests. Footnotes Figure 1, Figure 2 and Supplemental Figure S1 were revised. The manuscript and supplementary information were also revised.