Fig. 8

Genome-wide predictions of active CRMs in a human cell/tissue type and their reutilizations in different cell/tissue types. A Number of active CRMs predicted by the UFSP (active CRMs) versus the size of the positive set CRM+TF+ (positive CRMs) in a cell/tissue type. B Boxplots of gene expression levels in a cell/tissue type showing that genes closest to the CRM+TF⁺ set (positive) or to the active CRMs predicted by the UFSP model but missed by dePCRM2 (active CRMs-UFSP) have significantly higher expression levels than genes closest to the predicted non-active CRMs (p<2.23×10⁻³⁰², Mann-Whitney U test). C Number of predicted active CRMs shared by different numbers of cell/tissue types. D Number of closest genes to the uniquely active CRMs shared by different numbers of cell/tissue types. E Distributions of phyloP scores of all-shared active CRMs, uniquely active CRMs, all active CRMs, and all non-active CRMs in the cell/tissue types. All distributions are significantly different from one another, p<2.23X10⁻³⁰² (K-S test). F The levels of shared active CRMs reflect lineage relationships of the cell/tissue types. Cell/tissue types were clustered based on the Jaccard index of predicted active CRMs in each pair of the cell/tissue types