(‘dyadic’). We arbitrate on these regions by first clustering them (using the methods in the following section) with an expected cluster size of 10,000 regions, and then for each cluster calculating (a) the mean posterior probabilities for promoter and enhancer calls separately, and (b) the mean number of reference epigenomes in which regions were called promoter or enhancer. Clusters of regions for which the differences in mean posterior probabilities (a) is smaller than 0.05, or for which the absolute log2-ratio of the number of epigenomes called as promoter or enhancer (b) is smaller than 0.05 are called true ‘dyadic’ regions, along with a small number of ‘ambiguous’ regions in state BivFlnk. Note that this particular clustering is only to arbitrate on these regions using group statistics instead of one-by-one; the final clusterings are described next. Overall, we define ~2.3M putative enhancer regions (12.63% of genome), ~80k promoter regions (1.44% of genome) and ~130k dyadic regions (0.99% of genome), showing either promoter or enhancer signatures across epigenomes.
