Whole Genome Bisulfite Sequencing (WGBS) Benchmark Data

A central question in the analysis of whole genome bisulfite sequencing (WGBS) data is to detect regions (collections of neighboring CpGs) with systematic differences between conditions, as compared to within-condition variability. These so-called Differentially Methylated Regions (DMRs) are thought to be more informative than single CpGs in terms of of biological function. The method dmrseq (described in the preprint at https://doi.org/10.1101/183210) provides a rigorous permutation-based approach to detect and perform inference for differential methylation by use of generalized least squares models that account for inter-individual and inter-CpG variability to generate region-level statistics that can be comparable across the genome. The framework performs well even on samples as small as two per group. To evaluate this method, a simulated benchmark dataset was created to mimic the properties of real WGBS data. This was done by splitting a set of biological replicates (controls) into two artificial groups and adding methylation differences to randomly selected regions. Differences are added between the first half of samples and the second half of samples. The simulated data is saved as an `.rds` file for analysis with R, which contains a bsseq object as well as a Granges object denoting the true DMRs. The output is also saved in the form of compressed tab-delimited text files (`.txt.gz`) for each sample containing methylated and unmethylated counts per CpG, in addition to a `.bed` file containing the true DMRs. The control data is downloaded from GEO (GSE64177), and 3000 DMRs are added using the `simDMRs` function of the dmrseq R package (available here: https://github.com/kdkorthauer/dmrseq). More details and a script to reproduce the generation of these files is available at the companion repository: https://github.com/kdkorthauer/dmrseqPaper.