Complete deconvolution of DNA methylation signals from complex tissues: a geometric approach

Area of Science:

• Epigenetics
• Computational Biology
• Bioinformatics

Background:

• Epigenetics research often analyzes DNA methylation in mixed tissue samples.
• Accurate estimation of cell composition is vital for interpreting these results.
• Existing methods for cell composition quantification have limitations due to reliance on prior information.

Purpose of the Study:

• To develop a novel deconvolution method for estimating cell compositions from DNA methylation data.
• To create a comprehensive pipeline for identifying cell types, proportions, and cell-type-specific CpG sites.
• To enable cell type assignment even with partial or no reference panel.

Main Methods:

• Developed Tsisal, a complete deconvolution method for DNA methylation data.
• Tsisal estimates the number of cell types, their proportions, and identifies cell-type-specific CpG sites.
• The method can also assign cell type labels when reference panels are available.

Main Results:

• Tsisal accurately estimates cell compositions without requiring prior knowledge of cell types or proportions.
• The method demonstrated favorable performance in extensive simulations and analyses of seven real datasets.
• Tsisal outperformed existing deconvolution methods in similar applications.

Conclusions:

• Tsisal offers a robust solution for cell deconvolution in DNA methylation studies.
• The method enhances the accuracy of epigenetic analyses by accounting for cellular heterogeneity.
• Tsisal is available as an R/Bioconductor package (TOAST) for broader accessibility.