Using beta-binomial regression for high-precision differential methylation analysis in multifactor whole-genome bisulfite sequencing experiments

Background: Whole-genome bisulfite sequencing currently provides the highest-precision view of the epigenome, with quantitative information about populations of cells down to single nucleotide resolution. Several studies have demonstrated the value of this precision: meaningful features that correlate strongly with biological functions can be found associated with only a few CpG sites. Understanding the role of DNA methylation, and more broadly the role of DNA accessibility, requires that methylation differences between populations of cells are identified with extreme precision and in complex experimental designs. Results: In this work we investigated the use of beta-binomial regression as a general approach for modeling whole-genome bisulfite data to identify differentially methylated sites and genomic intervals. Conclusions: The regression-based analysis can handle medium-and large-scale experiments where it becomes critical to accurately model variation in methylation levels between replicates and account for influence of various experimental factors like cell types or batch effects.