Skip to main content


Fig. 3 | BMC Genomics

Fig. 3

From: Genome-wide enhancer annotations differ significantly in genomic distribution, evolution, and function

Fig. 3

Enhancer sets have low genomic overlap. (a) Pairwise bp enrichment values (log2 fold change) for overlap between each K562 (upper triangle) or liver (lower triangle) enhancer set, compared to the expected overlap between randomly distributed, length-matched regions. (b) The log2 enrichment for bp overlap compared to a random genomic distribution for each pair of enhancer sets within each context. Only contexts with annotations across all biological contexts are included. The fold changes across annotations for the primary tissues—liver and heart—are significantly lower than the cell lines—K562 and Gm12878 (p = 6.88E-11 Kruskal-Wallis test, followed by Dunn’s test with Bonferroni correction for pairwise comparisons between contexts). The patterns are similar for element-wise comparisons (Figs. S3). (c, d) The percent base pair (bp) overlap between all pairs of (c) K562 enhancer sets and (d) liver enhancer sets. Percent overlap for each pair was calculated by dividing the number of shared bp between the two sets by the total number of base pairs of the set on the y-axis. The highest overlap is observed for pairs based on similar input, e.g., machine learning models trained on the same functional genomics data, or comparisons with large sets, e.g. ChromHMM. Comparisons between biological replicates average 76% overlap. (e, f) The Jaccard similarity between all pairs of (E) K562 or (f) liver enhancer sets. The upper triangle gives the Jaccard similarity, and the lower triangle gives the relative Jaccard similarity in which the observed similarity is divided by the maximum possible similarity for the pair of sets

Back to article page