In Fig. 3a each CG-DMR was divided into 20 equally sized bins. The average methylation for all cytosines in the CG context within a bin in one sample was determined and normalized by the bin size. Lastly, the whole data set was divided by its 70th percentile, and values higher then 1 were forced to 1. This was performed to produce a meaningful mapping between values and colours in the heatmap key, and to avoid extreme values masking the methylation levels of other CG-DMRs. CG-DMRs were then reorganized based on their similarity by means of complete linkage hierarchical clustering, using the heatmap.2 R function.
