MethylBench: A comprehensive benchmark of DNA methylation profiling methods across diverse sequencing platforms
BackgroundDNA methylation can be profiled using multiple technologies that vary in resolution, coverage and cost. Prior benchmarking efforts have laid important groundwork, yet critical gaps remain. The SEQC2 EpiQC study[14] provided a multi-platform QC framework across whole-genome bisulfite sequencing (WGBS), oxidative bisulfite sequencing, enzymatic deamination, ONT and Illumina 850k arrays, showing high overall concordance but limited replication and no coverage of current enzymatic conversion, hybridization-based panels or modern long-read chemistries. More recently, Sigurpalsdottir et al.[40] systematically compared methylation detection tools within the long-read domain, demonstrating high ONT accuracy relative to oxidative bisulfite sequencing, but restricted their analysis to ONT and PacBio, excluding short-read, targeted-panel and array-based platforms as well as the biological interpretability of differential methylation across platform classes. A comprehensive cross-platform benchmark addressing coverage heterogeneity and annotation redundancy in the interpretation of differentially methylated cytosines thus remains absent. MethodsWe compared six widely used technologies - Illumina EPIC array, TWIST, Whole-Genome Enzymatic Conversion (WGEC), Reduced Representation Bisulfite Sequencing (RRBS), Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT) - using GIAB reference samples and ten blood/fibroblast samples from 5 individuals. We assessed CpG coverage, consistency of differentially methylated cytosine (DMC) detection, and genomic annotation, with particular attention to overlapping signals across assays. ResultsDespite major assay differences, all technologies consistently identified DMCs enriched in promoter and intronic regions, marking these as robust hotspots of epigenetic variability. Annotation redundancy strongly shaped initial interpretations, with CpG island-related categories largely disappearing once annotations were collapsed to unique features. Sequencing-based methods (WGEC, TWIST, ONT) achieved the most comprehensive coverage, while EPIC arrays reliably captured promoter-associated differences despite limited scope. ONT showed strong concordance with short-read methods after coverage filtering, but required higher, more uniform coverage for reproducible CpG-level agreement. PacBio showed a coverage-independent discrepancy, with concordance plateauing regardless of mean coverage - pointing to residual technology-specific bias rather than a simple coverage effect. ConclusionsCross-platform benchmarking yields coherent biological insights once coverage and annotation redundancies are addressed. EPIC arrays remain valuable for promoter-focused cohort studies, WGEC and TWIST enable genome-wide discovery and ONT offers unique phasing and multimodal potential - together guiding method selection and more robust interpretation of DNA methylation data.