Fig. 4

MDL plots as a signature of methylome. a Unsupervised clustering of MDL plots. WGBS data for human fibroblast IMR90 and ESC H1 [4], ESC-derived lineages including mesendoderm, trophoblast-like cell, neural progenitor cell and mesenchymal stem cell [6], cultured mammary epithelial cells and a low-passage breast cancer cell line [18], hematopoietic lineage cells including hematopoietic stem/progenitor cell (Hspc), CD133-positive hematopoietic stem cell (CD133 hsc), neutrophil and B-cell [19], placenta [16] and sperm [24] were subjected to changepoint detection followed by MDL plotting. We also included variously down-sized IMR90 data (see below). CpG sites covered by at least five reads were used for calculation. We performed Ward’s clustering using the Spearman’s rank-order correlation coefficient (see Methods). The heatmap illustrates the relationships across cell types based on MDL plots. The graded colors from red to white at the top left represent from similar to dissimilar in terms of the Spearman’s rank-order correlation coefficient between samples. b Effects of data size on the number of domains. We segmented the methylomes using the ten variously downsized IMR90 data under three different thresholds or minimal depth to select CpG sites for calculation of methylation level (i.e., ≥3, ≥5 or ≥10 reads). c Effects of data size on genomic coverage. d Effects of data size on MDL plot of IMR90 cells. The percentile on each plot indicates the fraction of data used for the analysis. CpG sites covered by at least three reads were used for calculation