Fig. 1

Human reference UMRs. a The statistical framework for the identification of conserved reference UMRs using WGBS data from 30 normal tissues and 35 solid tumors (Online Methods). b Cumulative distribution of UMR width for normal and tumor samples. Reference canyons (length > 3.5 kb) account for about 6% of all reference UMRs. The remaining short UMRs are regarded as control cUMRs. c Percentage of reference canyons/cUMRs covered by CpG islands (downloaded from UCSC). For each reference UMR, the percentage of UMR covered by the CpG islands is defined as the length of the UMR covered by CpG islands (either partially or entirely) divided by the total length of the UMR. Random represents 46,384 randomly selected regions for human genome that have the same length distribution, but without overlapping with reference UMRs. d Percentage of reference canyons/cUMRs covered by DNase I hypersensitivity clusters (DNaseI cluster) 125 cell types [23], transcription factor binding site clusters (TFBS cluster) of 161 TFs in 91 cell types [24], and enhancer clusters of H3K27ac peak regions in 88 human cell types [25]