cyberDMR: accurate and robust identification of differentially methylated regions from WGS-derived methylomes

Differentially methylated regions (DMRs) are essential regulatory features for characterizing epigenetic variation, understanding population-level differences, and interpreting haplotype-specific methylation patterns. However, existing DMR detection methods often show unstable performance under biologically challenging scenarios such as sparse CpG regions, low sequencing depth, small sample sizes, extreme region lengths, and subtle methylation differences. These limitations highlight the need for a more robust and versatile approach capable of achieving accurate and consistent DMR detection across diverse genomic contexts. To address these challenges, we developed cyberDMR, a robust and noise-resilient DMR detection framework that integrates coverage-adaptive smoothing, seed-growing clustering, and weighted beta regression to achieve stable performance under whole-genome sequencing (WGS) and varying sequencing conditions.

cyberDMR consistently achieved the highest F1-scores on real WGBS prostate cancer datasets, outperforming other strong-performing methods by 2-18%. This advantage was further supported by extensive simulations spanning diverse genomic and experimental scenarios, in which cyberDMR maintained stable and leading performance, achieving a mean F1-score of 99.7% (± 0.3%) and consistently outperforming competing methods across diverse simulated settings. When applied to long-read methylation datasets, cyberDMR enabled the identification of 107 biologically meaningful methylation regions, including 15 gene-associated regions influenced by cis-acting methylation quantitative trait loci (meQTLs) and 2 corresponding putative novel imprinting candidates, ZNF714 and NINJ2.

cyberDMR provides a robust and accurate framework for genome-wide DMR detection, capturing consistent inter-group methylation differences while preserving intra-group consistency across diverse sequencing conditions.