Long-read platforms support large DNA methylation predictors

Using human blood samples from two population cohorts, researchers evaluated how long-read sequencing platforms capture DNA methylation compared with standard Illumina EPIC arrays. The team analyzed samples from the PEGS study and 1,815 long-read datasets from 1,787 All of Us participants using Oxford Nanopore and PacBio HiFi technologies. Both platforms measured roughly 28 million CpGs at 10× depth or higher. Between paired Nanopore and PacBio assays, overall Pearson correlations reached 0.85 to 0.87, but mean-centered correlations dropped, indicating limited agreement at single CpGs. In simulations, read depths of 50× to 60× were needed to match array-level technical reproducibility. When evaluating 173 methylation-based predictors, model size dictated cross-platform portability: 19 of 21 predictors with more than 1,000 CpGs achieved correlations above 0.8 between platforms, compared with only four of 137 smaller predictors.