Extended Data Fig. 17: Enhancer target gene predictions.

a, Schematic of the approach to assign enhancers to target genes. b, Genome browser view showing the Ascl1 locus, as in Fig. 4c, but showing ChIP–seq fold enrichment tracks instead of chromatin states. c, Histogram of the number of enhancers per gene. d, For each replicate, the fraction of putative enhancers assigned to the same gene using data from the other available replicate. e, Scatter plots showing reproducibility of enhancer–gene maps as measured by correlation between enhancer–gene pairs (left; n = 21,141 pairs), and the number of enhancers per gene (right; n = 5,611 genes). f, Left, fraction of enhancer–gene associations that overlap interactions previously reported in ref. ⁶ (n = 907/12,655), GeneHancer⁸¹ (n = 2,067/12,546), JEME⁸² (662/36,007), and RIPPLE⁸³ (31/37,541). The global level of overlap is low, perhaps in part owing to the different sample types used to predict these interactions. Right, distribution of scores for the unique and overlapping pairs in GeneHancer, JEME and RIPPLE, respectively. Where predictions from those reports overlap with ours, their scores are significantly higher. P values calculated using two-sided Mann–Whitney U test. g, As in Fig. 4d, this plot shows that enhancer–gene interactions identified by this correlative approach are generally more likely to be supported by chromatin interaction data than associations derived by a nearest gene approach. To ensure that this was not due to an artefact of the chromatin capture technologies being unable to detect short-range interactions, we used different distance cutoffs (10 kb, 100 kb) to define the ‘nearest’ non-target gene. h, The Bcl11a locus (chr11: 24,044,043–24,197,927; mm10) provides an interesting case in which genetic variation in enhancers regulating a pleiotropic Mendelian disease gene may contribute to tissue-restricted phenotypes with lower penetrance. Boxes outline enhancer clusters with active chromatin signatures in the CNS (left) and liver (right), and which have validated activity in the CNS and erythroid lineage, respectively^34,40,97 (mouse embryonic liver is a site of erythropoiesis). The subpanels on either side of the main browser view show regions of the human genome that correspond to either the CNS enhancer cluster (left, chr2: 60,752,530–60,767,198; hg19) or liver enhancer cluster (right, chr2: 60,711,940–60,741,118; hg19). Thick black bars on top represent orthologues of the predicted Bcl11a enhancers, and thin green bars below represent GWAS SNPs for the EMBL-EBI GWAS catalogue.