New BLEND-M Method Uses Personalized DNA Methylation References for Deconvolution

Researchers have introduced a novel computational method called BLEND-M, designed to improve the accuracy of cellular deconvolution in bulk DNA methylation data. Published in Clinical Epigenetics, the study addresses a significant limitation in current genomic analysis techniques: the assumption that cell type-specific DNA methylation profiles are consistent across all individuals. By learning personalized reference profiles for each bulk sample, the new approach aims to provide more precise estimates of cell type fractions, which are critical for various downstream genomic analyses.

Cellular deconvolution serves as an in-silico alternative to traditional cell sorting techniques, which are often costly and labor-intensive. The process estimates the proportions of different cell types present in a tissue sample. Accurate estimation of these fractions is essential for DNA methylation analyses because it allows researchers to adjust for confounding variables, support risk prediction models, and facilitate cell type-specific differential methylation studies.

A key determinant of deconvolution accuracy is the choice of cell type-specific reference profiles, which represent the expected DNA methylation levels in each cell type. It is widely recognized that these methylation levels vary substantially across individuals. However, existing methods typically assume a shared reference profile across the entire population. This assumption can lead to biased fraction estimates, as it fails to account for the natural inter-individual variability in DNA methylation patterns that occurs within a population.

To address this issue, the authors developed BLEND-M, a method that learns personalized cell type-specific reference profiles for each bulk sample. This approach explicitly accounts for the variability in DNA methylation levels between individuals. Furthermore, BLEND-M models heteroscedasticity across marker CpGs, acknowledging that DNA methylation measurements at some CpGs are noisier than others.

The study establishes theoretical guarantees for the accuracy of the BLEND-M method. The authors demonstrate its improved performance relative to existing methods through extensive benchmarking on both realistic simulated and real datasets.

Beyond its technical improvements, the study illustrates the practical utility of BLEND-M by applying it to develop a risk prediction model for childhood atopic asthma. This application demonstrates how more accurate deconvolution can enhance downstream analyses, including risk prediction.

The advances offered by BLEND-M enhance a range of downstream analyses, including confounding adjustment and cell type-specific differential methylation studies.

The study, authored by Penghui Huang, David G. Peters, and Chris McKennan, was published in Clinical Epigenetics in August 2026.

Subject of Research: Computational method for DNA methylation deconvolution using personalized references

Article Title: BLEND-M: cellular deconvolution with personalized DNA methylation references

Article References: Huang, P., Peters, D. G., & McKennan, C. (2026). BLEND-M: cellular deconvolution with personalized DNA methylation references. Clinical Epigenetics, 18(1), Article 205. https://doi.org/10.1186/s13148-026-02223-9

Image Credits: AI Generated

DOI: 10.1186/s13148-026-02223-9

Keywords: DNA methylation, cellular deconvolution, BLEND-M, genomics, cell type fractions, personalized references, Clinical Epigenetics, risk prediction, cellular, deconvolution, personalized, methylation