ROSE called super-enhancers overlapping with promoters

Question

[os306]

Hello,

I have used ROSE to call super-enhancers in my H3K27ac ChIP-seq data. I have used default settings (i.e. a stitching distance of 12,500 bp and a TSS exclusion zone size of +/- 2,500 bp. I have also used ChIP-seeker to annotate my H3K27ac peaks, and have defined promoters -2500 bp to +2500 bp from the TSS. With these settings specified, I was expecting to see no overlap with my ROSE-called SE regions and my ChIP-seeker promoter regions. However, when I load the .bed files into IGV, I noticed that some of my SE regions contain TSS and ChIP-seeker annotated promoter peaks (see attached figure). I guess my questions are as follows:

1. Why were these regions called super-enhancers by ROSE - I thought the specification of a TSS exclusion zone would mean that none of the SEs would overlap with promoter regions
2. Would it make biological sense to use bedtools to subtract the genomic coordinates of my ROSE-called SEs from the list of genomic coordinates for my promoters (to ensure there is no overlap between the two. Or should I do things vice versa (i.e. remove any SEs that overlap with promoters from my SE list)?

I know these may be pretty silly questions, but I haven't been able to find a good answer to this anywhere else. Would be very grateful for any help!

Answer 1

[jared.andrews07]

Why were these regions called super-enhancers by ROSE - I thought the specification of a TSS exclusion zone would mean that none of the SEs would overlap with promoter regions

It only excludes peaks that overlap the TSS regions from stitching. It doesn't mean the TSS itself can't be included in an SE, e.g. if you have peaks bookending the promoter that are <12.5kb apart, they'll be stitched and include the TSS. There is also an undocumented behavior whereby a region spanning TSSes from 3+ genes will not be stitched together. This means that gene dense regions tend to have fewer SEs called when by eye they look like pretty obvious SEs. Also, the annotations built into ROSE are super old and definitely missing certain genes you'd likely expect. If you care much, I'd roll your own and edit the source code to use them. ROSE is really showing its age and could desperately use some updates.

Would it make biological sense to use bedtools to subtract the genomic coordinates of my ROSE-called SEs from the list of genomic coordinates for my promoters (to ensure there is no overlap between the two. Or should I do things vice versa (i.e. remove any SEs that overlap with promoters from my SE list)?

If you remove SEs that overlap with promoters, you're going to remove a lot of SEs. Potentially a majority of them. These regions overlap, they need not be mutually exclusive (as promoters can certainly act as enhancers).

[os306]

Thank you very much, I really appreciate your reply!

[os306]

Out of interest, what annotation would you recommend using instead of the default built-in ones. Thanks again!

[jared.andrews07]

The most recent one for your organism from Ensembl or Gencode, optionally filtered to remove genes with less evidence (e.g. the "basic" Gencode annotations).