We estimated enhancer activity of candidate elements using a combination of quantitative DNase-seq and H3K27ac ChIP-seq signals. DNase accessibility and acetylation of H3K27 are commonly used to identify enhancer elements44 ,45 , and are predictive of the expression of nearby genes and enhancer activity in plasmid based reporter assays46 –48 . Quantile normalization of epigenetic signals is used to facilitate comparison of ABC Scores across cell types (see Supplementary Methods).
DNase peaks were extended 175 bp because H3K27ac ChIP-seq signals are strongest on the nucleosomes flanking the nucleosome-free DHS peak. We computed the geometric mean of DNase-seq and H3K27ac ChIP-seq signals because we expect that strong enhancers should have strong signals for both, and that elements that have only one or the other likely represent other types of elements. (Elements with strong DNase-seq signal but no H3K27ac ChIP-seq signal might be CTCF-bound topological elements. Elements with strong H3K27ac signal but no DNase-seq signal might be sequences that are close to strong enhancers but do not themselves have enhancer activity, due to the spreading H3K27ac signal over hundreds to thousands of bp.) We report sources of epigenomic data in Supplementary Table 4. Where replicate experiments are listed, we averaged the signal in each element across the replicates unless otherwise stated.
We note that this calculation of enhancer activity is the same for a given element across all genes. This means that the model assumes that an enhancer has the same “Activity” for every promoter (i.e., no differences due to biochemical specificity).