Methylation map covers 28 million CpG sites — and a model predicts how variants shift them

DNA methylation — a chemical tag on cytosine that helps decide which genes are active — is central to development, imprinting and disease, yet most large-scale studies still rely on arrays that read out only about 3% of the genome's roughly 28 million CpG sites. A computational framework described in Nature Genetics now offers whole-genome coverage. The work, led by Olga G. Troyanskaya at the Flatiron Institute of the Simons Foundation, with first author Weiguang Mao, combines a harmonized compendium of methylation data,
<truncated 2096 bytes>
CpG resolution for all 296 profiles at once, using a 2,048-base-pair sequence window. On a chromosome holdout set it reached an average Spearman correlation of 0.640 and Pearson correlation of 0.825 across profiles. The authors report that Wreath is the first model to predict continuous genome-wide methylation at single-CpG resolution across a broad range of tissues.

Against GTEx methylation quantitative trait loci, Wreath correctly predicted the direction of tissue-specific variant effects for more than 90% of the top 20% highest-impact variants, approaching 98% in blood — and matching tissues correlated better than mismatched ones.

Applied to 1,121 curated noncoding pathogenic variants from the Human Gene Mutation Database, predicted methylation effects were larger than those of common 1000 Genomes variants, with 56% predicted to increase methylation and 44% to decrease it. For example, variants linked to NAGS deficiency were predicted to raise methylation in liver and gallbladder, with attribution analysis pointing to disruption of an HNF-1 binding site; variants at the HBG2 promoter were predicted to lower blood methylation, offering a possible explanation for persistent fetal hemoglobin.

The compendium, methylation classes and model are being made available to the community. The obvious next step is experimental validation of the predicted variant effects, and the authors note that pairing Wreath with matched whole-genome sequencing and methylation data could eventually support personalized predictions.

For social sharing: A new framework maps DNA methylation across 28 million CpG sites and 90-plus tissues — then predicts how individual variants change it.

Reference: Mao, W., Chen, K.M., Park, C.Y., Theesfeld, C.L. & Troyanskaya, O.G. A framework for genome-wide characterization of DNA methylation and variant effect prediction. Nature Genetics (2026). https://doi.org/10.1038/s41588-026-02807-4