EMeth: An EM algorithm for cell type decomposition based on DNA methylation data.

Area of Science:

• Computational biology
• Epigenetics
• Bioinformatics

Background:

• Estimating cell type proportions from bulk tissue DNA methylation data is crucial for understanding tissue composition and disease mechanisms.
• Existing reference-based methods face challenges with data heterogeneity and accurately deconvoluting complex cellular mixtures.

Purpose of the Study:

• To introduce EMeth, a novel reference-based computational method for improved cell type proportion estimation using DNA methylation data.
• To address limitations in existing deconvolution methods by incorporating CpG-level quality control and enabling estimation of unknown cell type-specific methylation profiles.

Main Methods:

• EMeth employs a reference-based approach, utilizing cell type-specific DNA methylation data.
• It identifies and down-weights CpG sites with methylation levels inconsistent with the deconvolution model, enhancing robustness.
• A novel feature allows for the estimation of methylation profiles for cell types with known proportions but unknown references, applicable in tumor microenvironment studies.

Main Results:

• EMeth demonstrated superior accuracy in estimating cell type proportions compared to existing methods, validated using simulated and in silico mixture data.
• Application in cancer studies revealed that EMeth-derived T regulatory cell proportions showed expected correlations with mutation load and survival, associations missed by gene expression-based estimates.

Conclusions:

• EMeth provides a more accurate and robust method for cell type deconvolution from DNA methylation data.
• The method's ability to handle unknown references and its improved accuracy offer significant advantages for cancer research and other complex biological systems.