If you look at the "GeneAnno" plot (ie genomic region annotations), "genic enhancers" occur more frequently in gene bodies, and occur in exons, while "enhancers" (state 7) occur less frequently in gene bodies and almost never in exons. Both seem to occur in introns with equal frequency. If you dig into the raw data, you can also see that the "genic enhancers" occur at TSS about twice as frequently. Also, the "genic enhancers" occur about 26 times less frequently genome-wide.

http://egg2.wustl.edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/coreMarks/jointModel/final/annotationEnrichment_RoadmapEp_coreMarks_15State.png

http://egg2.wustl.edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/coreMarks/jointModel/final/E003_15_coreMarks_geneAnno_overlap.txt

It's important to note that these are probabilistic categories based solely on the histone marks -- there are probably some promoters/TSS mixed in with the regions annotated as genic enhancers. Conversely, many of the genic enhancers may be located intergenically. Lastly, there is no guarantee that any of the regions annotated as enhancers actually function as such. If they do function as enhancers, then they must, by necessity, be brought into proximity to promoters via enhancer-promoter looping. This mechanism functions regardless of whether they are genic or intergenic.

Can we generate consensus plots from multiple samples of each subtype using Chromhmm?