Bridging the gap: Enhancing the generalizability of epigenetic clocks through transfer learning

Changes in DNA methylation patterns exhibit a high correlation with chronological age. Epigenetic clocks, developed through statistical models that estimate epigenetic age using the methylation levels of cytosine-guanine dinucleotide (CpG) sites, have emerged as powerful tools in understanding aging and age-related diseases. Despite their popularity, the generalizability of these clocks across diverse populations remains a challenge. We find that some of the widely used epigenetic clocks, such as Horvath’s clock (Horvath, 2013) and PedBE clock (McEwen et al., 2020) do not perform well in our target cohort. This lack of representativeness raises concerns about applying these clocks to quantify biological age in distinct demographic and ethnic groups. In addition, the feature space between existing clocks and our target data is different: most existing clocks are trained with data from older platforms, such as the Illumina HumanMethylation450 BeadChip (450K). In contrast, our target data are profiled with a more recent Illumina HumanMethylationEPIC BeadChip (EPIC) array. To address these gaps, we propose a transfer learning framework to adapt existing epigenetic clocks to underrepresented populations, using shared knowledge from diverse datasets. Furthermore, we develop imputation- and DNN-based methods for feature adaptation between existing clocks and our target data. Using data collected from 593 blood samples from a cohort of children and adolescents in the ELEMENT study, we find that our proposed transfer learning methods greatly improve the prediction performance compared to applying existing clocks directly. Performance is further enhanced by using the CpG sites profiled on the EPIC array. Our methodology showcases the potential to bridge the gap between different DNAm datasets and different profiling platforms, thus improving the applicability of epigenetic clocks in diverse population groups and contributing to more accurate aging research.