Peak to gene links - restrict to TSS links?

[jblich870]

Hi,

I've run addPeak2GeneLinks followed by getPeak2GeneLinks with returnLoops = T, and have attached a representative track below.

Plot.pdf

Many of the displayed 'peak2genelinks' do not include the TSS - instead of linking peaks to genes, they appear to be just linking peaks.

Why is this? And is there a way to restrict the links so they only link peaks to genes?

Many thanks,
J

[rcorces]

This shouldnt be the case. Can you provide a reproducible example using the tutorial data? My assumption is that there is something off about your geneAnnotation

[jblich870]

I'm using ArchR-provided hg38:

addArchRGenome("hg38")

Using your code for my project gives the below. Seems odd that there are 166023 TSSs, versus only 49052 with hg19?

gA <- getGeneAnnotation(Multiproj)
gA
List of length 3
names(3): genes exons TSS

gA$TSS
GRanges object with 166023 ranges and 2 metadata columns:
seqnames ranges strand | tx_id tx_name
|
[1] chr1 29554 + | 1 uc057aty.1
[2] chr1 30267 + | 2 uc057atz.1
[3] chr1 30366 + | 3 uc031tlb.1
[4] chr1 69091 + | 4 uc001aal.1
[5] chr1 160446 + | 5 uc057aum.1
... ... ... ... . ... ...
[166019] chrY 25394719 - | 182394 uc065cwf.1
[166020] chrY 25391610 - | 182395 uc065cwg.1
[166021] chrY 25498059 - | 182396 uc065cwl.1
[166022] chrY 26411158 - | 182397 uc065cwp.1
[166023] chrY 57203357 - | 182398 uc065cww.1

---

seqinfo: 24 sequences from hg38 genome

[rcorces]

This is expected as genome annotations continue to improve over time. hg38 has better transcript annotations.

I think the point that hasnt been conveyed is that one gene can have many transcriptional start sites. In the hg38 TxDb object, these alternative start sites are annotated and thus they are a part of the ArchRProject. In ArchR, only one gene is plotted in the browser track for cleanliness but there may be many transcripts for that gene, some of which might have alternative TSSs. So just because you dont see the different gene tracks for different transcripts, doesnt mean that they dont exist. Perhaps this is what you were observing previously when you said that your links didnt map to TSSs.

[jblich870]

Right, I hadn’t considered that - that helps to clear things up a bit, thanks.

It does however raise a couple of separate, but related, issues:

First is that some genes seem to be absent from the browser track whereas their TSSs are annotated in the hg38 object. This leads to the peculiar situation of having links in the track for invisible genes (e.g. AC016745.3 in the attached plots).

Plot 5.pdf

Second is that some links appear to be between the promoter of one gene and the TSS of a neighbouring gene (eg below). This is probably because genes in close proximity on the genome can covary in expression, and so size of the promoter will correlate with expression of the neighbouring gene even though there is unlikely to be a biological link between the two. Is there a way to filter out such links?

Plot 7.pdf

Additionally, it might be helpful to be able to selectively visualise links for only one gene / one TSS in the browser track. Many links are long range, but in order to capture these in the track you end up getting lots of links for nearby genes and the plot can look messy.

[rcorces]

My general response is: Gene and transcript annotation is a complicated problem that we (ArchR) are not wading into and we have entirely off-loaded this problem to the end user and TxDb. ArchR will use whatever gene annotation you give it. If you dont like your gene annotation, change it to your liking. But ArchR isnt a gene annotation package so we steer clear of this entirely.

First is that some genes seem to be absent from the browser track whereas their TSSs are annotated in the hg38 object. This leads to the peculiar situation of having links in the track for invisible genes (e.g. AC016745.3 in the attached plots).

As mentioned above, this falls into the "if you dont like the default, make your own".

Second is that some links appear to be between the promoter of one gene and the TSS of a neighbouring gene (eg below). This is probably because genes in close proximity on the genome can covary in expression, and so size of the promoter will correlate with expression of the neighbouring gene even though there is unlikely to be a biological link between the two.

Its all about correlation. There is no true implication of causation and there are examples in biology of the promoter of one gene "regulating" the expression of another gene.

Is there a way to filter out such links?

Sure, but this is up to the end-user. Filter the returned GRanges object however you want.

Additionally, it might be helpful to be able to selectively visualise links for only one gene / one TSS in the browser track. Many links are long range, but in order to capture these in the track you end up getting lots of links for nearby genes and the plot can look messy.

You can add arbitrary tracks to your browser visualizations. You can make whatever GRanges object you want by filtering as mentioned above and add it as a loopTrack to plotBrowserTrack() or specify the loops argument to ArchRBrowser().

[jblich870]

Fair enough, and good points/suggestions.
Really appreciate your help, thanks.
J