Differential methylation analysis for BS-seq data under general experimental design

DNA methylation is an epigenetic modification with important roles in many biological processes and diseases. Bisulfite sequencing (BS-seq) has emerged recently as the technology of choice to profile DNA methylation because of its accuracy, genome coverage and higher resolution. Current statistical methods to identify differential methylation mainly focus on comparing two treatment groups. With an increasing number of experiments performed under a general and multiple-factor design, particularly in reduced representation bisulfite sequencing, there is a need to develop more flexible, powerful and computationally efficient methods. We present a novel statistical model to detect differentially methylated loci from BS-seq data under general experimental design, based on a beta-binomial regression model with 'arcsine' link function. Parameter estimation is based on transformed data with generalized least square approach without relying on iterative algorithm. Simulation and real data analyses demonstrate that our method is accurate, powerful, robust and computationally efficient. It is available as Bioconductor package DSS. yongpark@pitt.edu or hao.wu@emory.edu Supplementary data are available at Bioinformatics online.