Question: Change in pathway IDs scheme in reactome.db
1
2.5 years ago by
assaron150
assaron150 wrote:

I just found out that in recent reactome.db version (1.59.0) there was a change of pathways IDs (PATHID column). Before there was 20800 pathways with pathways beeing specific to an organism. Now there are 2185 pathways, shared between organisms. I'm not sure what were the reasons for the change, but I don't have problems with both schemes. However, I find it inconsistent, that for a single pathway ID there are multiple pathway names.

With reactome.db 1.58.0:

> length(keys(reactome.db, keytype="PATHID")) [1] 20800 > AnnotationDbi::select(reactome.db, keys=c("Bos taurus: Interleukin-6 signaling", "Homo sapiens: Interleukin-6 signaling"), columns=c("PATHID"), keytype=c("PATHNAME")) 'select()' returned 1:1 mapping between keys and columns                                PATHNAME  PATHID 1   Bos taurus: Interleukin-6 signaling 5870529 2 Homo sapiens: Interleukin-6 signaling 1059683

With reactome.db 1.59.0:

> length(keys(reactome.db, keytype="PATHID")) [1] 2185 > AnnotationDbi::select(reactome.db, keys=c("Bos taurus: Interleukin-6 signaling", "Homo sapiens: Interleukin-6 signaling"), columns=c("PATHID"), keytype=c("PATHNAME")) 'select()' returned 1:1 mapping between keys and columns                                PATHNAME  PATHID 1   Bos taurus: Interleukin-6 signaling 1059683 2 Homo sapiens: Interleukin-6 signaling 1059683

As this behavior breaks build of my package (fgsea) and I need to fix it, I wonder is this transition final? If yes, may be it's better to remove organism specific prefix whatsoever? I'm pretty sure it's much better to have one-to-one map between pathway IDs and pathway names.

Best,
Alexey

reactome.db fgsea • 940 views
modified 2.5 years ago by willem.ligtenberg150 • written 2.5 years ago by assaron150
Answer: Change in pathway IDs scheme in reactome.db
0
2.5 years ago by
Netherlands
willem.ligtenberg150 wrote:

No that change is not final.
I was alerted to that issue last weekend. I have already created a new version which results in a one-to-one mapping of pathway-id to pathway name.

Apparently, Reactome has changed their identifiers, and I have revised the way I am building the package.
I mistakenly thought that the R-HSA-<ID> thing was only in their exports, but they have made that change throughout the database and website. (When I started using Reactome the DB_IDs were always numeric.)
So I removed the R-<three letter species>- bit from all ids, which then resulted in the multi mapping issue.

When I was alerted to this, I now have made a package that does not remove the R-HSA- part. So all IDs are now of the form: R-<three letter abbreviation species>-<number>.
I then had the request to also include the old (full numeric) ids that were present previously, but I can only do that with up to date info for human, which I think is not what people would expect, since there are also users using Reactome with  non-human data.
Another option would be to include the mappings from the previous version, which would mean that that data is stale. (Also sub-optimal)

Any suggestions on the preferred way forward are welcome. I can build another version relatively quickly.

If you want to test that new version feel free to download it here.

1

I think the form  R-<three letter abbreviation species>-<number> is good enough. Not sure why would anyone want the ids to be numeric only.

Backwards compatibility, if you have a report (knitr/sweave) that might break now, if you were using the database ids to specifically look at a pathway.
However, I do think that the main use case is starting at gene ids, finding related pathways (first pathway id, then the pathway name) that would still work normally.

I see. Well, as you say, the package just reflects reactome database, so I'd argue it's better to have ids consistent with current reactome ids, that is of the form like R-HSA-<ID>.

Btw, did I got it right that there is a new version coming very soon with unique ids?

1

Yes, you did get that right.
You can download it here if you want to test drive it:
https://owncloud.wligtenberg.nl/index.php/s/YfcW3XU19ptgrOR

When it should be available at bioconductor 3.5? It's a release date soon...

Did you verify this fixes your issue?
These annotation packages are not tracked in version control. If this fixes your issue, I will ask Valerie is she can upload the version that I have on my website.

Oh, I see. Yes, it does.

The annotation push script is run nightly so this should
appear in the repo tomorrow (April 21).

I still can't see the updated version in Bioconductor. Is this OK? There are still 2185 PATHID's in the packge. Also, should the reactome.db version change with this update, from 1.59.0 to 1.59.1 or something?

I didn't update the version number (maybe should have...)

1

Can you check with Valerie?  It'd be better with new version number, because it would be easier to check if it was updated or not.

1

I mailed Valerie an updated package with the version number increase.

mmm, version 60 of Reactome was released, should I run a quick update?