Tissue and primary cell samples mapped to ontology facets were clustered by complete linkage agglomerative hierarchical clustering based on Jensen-Shannon (JS) divergence59. In detail, expression values for all enhancers in the permissive set were normalized to sum to 1 for each sample and the square root (proper distance metric) of all pair wise JS divergences between samples was calculated. Manually selected clades of samples were analyzed for differential expression in a similar way as was done for facets (see above). In summary, differentially expressed enhancers (robust set) were identified by Kruskal-Wallis rank sum tests (Benjamini-Hochberg FDR<0.05) and subsequent NDWD post-hoc tests were performed to find all significant pair wise differences (Bonferroni single-step adjusted P<0.05) between clades.
