Addressing population and platform heterogeneity in epigenetic clocks via transfer learning and conformal prediction methods

/ ABSTRACT The objective of the proposed research is to develop a comprehensive framework to enhance the utility and accuracy of epigenetic clocks, which are tools used to predict chronological age or aging-related diseases from DNA methylation (DNAm) levels at specific cytosine-guanine dinucleotide (CpG) sites. Despite their widespread use, current epigenetic clocks suffer from the following issues: (i) reduced predictive power when applied to diverse study cohorts, (ii) lack of prediction interval that accounts for population heterogeneity, and (iii) insufficient use of data from higher-resolution platforms. To address these challenges, this proposal sets forth several aims. First, this project will develop a heterogeneity-aware transfer learning framework to improve the generalizability of existing epigenetic clocks to different populations, tissues, or cell types. This includes creating methods to correct biases and enhance prediction accuracy by leveraging information from established clocks while overcoming data-sharing constraints. Second, prediction intervals based on mean regression produce the same width across different subpopulations. This proposed work will establish a quantile regression framework to understand better and predict variations in age acceleration across different population segments. The key idea is to employ high- dimensional quantile regression and conformal inference to build adaptive prediction intervals. Furthermore, this proposed work will create an integrative method utilizing higher-resolution DNAm data to optimize the performance of epigenetic clocks originally developed with lower-resolution data. This method will use neural network transfer learning to integrate information across spatially correlated CpG sites. Lastly, the newly proposed methods will be implemented and disseminated to the scientific community with open-source software packages. The completion of this proposal will significantly improve the generalizability, adaptability, and transparency of epigenetic age calculators, enabling them to serve diverse populations better and incorporate data from newer, higher-resolution DNAm platforms. The availability of open-source software will also support further research and application of these improved models in aging research. Project Number: 1R01AG092615-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Institute on Aging (NIA) | Principal Investigator: Lan Luo | Institution: RUTGERS BIOMEDICAL AND HEALTH SCIENCES, Newark, NJ | Award Amount: $461,514 | Activity Code: R01 | Study Section: Analytics and Statistics for Population Research Panel A Study Section[ASPA] View on NIH RePORTER: https://reporter.nih.gov/project-details/11293237