Targeting methylated regions using covariation and distance

Abstract Background The analysis of WGBS (whole genome bisulfite sequencing) datasets is challenging. The large number of CpG sites requires significant computing power and can lead to harsh multiple correction penalties. Typically, the number of CpG sites found in DMRs (differentially regulated regions) represent a very small proportion of the initial number of CpG sites. This is because methylation levels of the majority of CpG sites do not vary significantly between samples, and/or the CpG sites are too far dispersed to be considered a contiguous region. DMRs are like likely to be found in relatively compact CpG rich regions that vary in methylation levels. Isolating these regions could greatly reduce downstream computational and statistical challenges without any previous knowledge of sample groups. Results The proposed method was able to isolate compact CpG rich variable regions using distance, covariation, and user parameters without a priori sample information. Results were verified with EpiDISH cell deconvolution and comparable with to a complementary method DMRSeq. Isolated regions averaged just 293 bp in length yet contained an average of 29 CpG sites per region. Conclusions By defining compact CpG rich variable regions, the method hopes to provide a valid and simpler starting point for further downstream analyses. This method is applicable to any dataset containing total CpG and total CpG methylated count matrices.