Fig. 10
Comparison of the performance of our method and Matched Filter for genome-wide predictions of active CRMs in six human cell lines. A Number of active CRMs predicted by the two methods in the cell lines. B Proportion of the human genome covered by active CRMs predicted by the two methods in the cell lines. C Venn diagram of the union of nucleotide positions covered by active CRMs predicted by each method in the six cell lines. D Distributions of phyloP scores of nucleotide positions covered by active CRMs predicted by both methods (shared) and by active CRMs predicted only by Matched Filter (Matched Filter only) or only by our method (our method only). E Boxplots of the expression levels of genes closest to active CRMs predicted by both methods (overlapping each other by at least 50% of the shorter one) in a cell line (shared) and of genes closest to active CRMs predicted only by Matched Filter (Matched Filter only) or only by our method (our method only). In each cell/tissue type, genes closest to active CRMs predicted by both methods and genes closest to active CRMs predicted only by our method have significantly higher expression levels than genes closest to active CRMs predicted only by Matched Filter (p<3.37×10−41, Mann-Whitney U test)