`standarize` gives misleading results on Arden style gene names

[JamieHeather]

Hi Yuta,

I was just using tidytcells to fix some TCR names from some older papers, when I noticed that it was getting a few of them wrong. When I looked into it they all seemed to be from older pre-IMGT nomenclatures, particularly that from Arden et al. I made use of these two papers with gene ID conversion tables:

• LeFranc et al. (which is basically the supplementary data following the original IMGT nomenclature approval)
• Arden et al. (which details that older nomenclature, handily providing a bunch of accessions, which are out of date but do link to modern sequence IDs for validation)

I then picked out a couple of example genes:

• TRAV1-2
  • In the Arden nomenclature this is AV7S2
  • (This is what actually made me start looking, when I noticed that some supposedly MAIT alpha chain antibodies bind 'TCRAV7S2')
• TRBV19
  • This is BV17S1 under Arden
  • (This is particularly relevant, as 'TCRBV17S1' is an example given in the tidytcells paper)

Running these through tidytcells gives the wrong answers:

import tidytcells as tt


tt.tr.standardize('AV7S2')
/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/tidytcells/_utils/warnings.py:12: UserWarning: Failed to standardize "AV7S2" for species homosapiens: unrecognised gene name. (best attempted fix: "TRAV7-2").
  warn(warning_message)

tt.tr.standardize('BV17S1')
'TRBV17'

I guess the issue is that Arden-style gene IDs look plausibly like old transitional IMGT IDs (which I think we've discussed in the past), and so they're getting simply updated, rather than looked up! It makes me wonder if there could/should be a way to stipulate whether a specific nomenclature is being used, or perhaps raise a flag or warning if an ambiguous older ID is used?

On a related note, I also noticed (while trying to standardise a bunch of V-gene specific antibodies) that it doesn't seem to handle the (presumably even older style) TCR names with Greek characters, e.g. using old TRAV1-2 protein names again:

tt.tr.standardize('Vα7.2')
/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/tidytcells/_utils/warnings.py:12: UserWarning: Failed to standardize "Vα7.2" for species homosapiens: unrecognised gene name. (best attempted fix: "TRV").
  warn(warning_message)

This makes sense, but given the retention of these names in product names it might be worth trying to cope with them in a future release.

[yutanagano]

Hi Jamie, thanks for taking the time to post the issue. I don't have much knowledge around how older gene symbols work so it's helpful that you are pointing out when things go wrong. 🙏🏼

First of all I agree that correct resolution of older gene names (or at least transparency of potentially incorrect or ambiguous standardisation) is important. tidytcells does have a rudimentary system to catch deprecated gene symbols, but it is currently just one dictionary look-up performed at the beginning (so even if tidytcells knows about "TCRAV7S2" it won't pick it up if it only says "AV7S2"). I should change this so that this synonym look-up is done at all levels so that resolution is more accurate.

The problem is more complicated though. I took a quick look at the two example genes that you pointed out, and I have questions for both.

TRAV1-2

I looked in tidytcells' synonym dictionary, and saw that it currently does not know about the symbol "TRAV7S2". I reviewed where tidytcells gets its TR synonym/deprecated symbols data from, and it's HGNC. Indeed, the HGNC entry for TRAV1-2 doesn't contain any info on synonyms or deprecated names. So as it stands, even after I update the synonym resolution logic, tidytcells won't know how to standardise "AV7S2". So how did you know that "AV7S2" refers to TRAV1-2? Is it just from the LeFranc et al paper that you linked? Asking in case you know of some other more complete / updated / nice programmatic source I can pull from.

TRBV19

The HGNC entry for TRBV19 does contain info on previous names, and marks "TCRBV17S1A1T" as a synonym. However "TCRBV17S1" is actually mapped to TRBV17. In your case, it seems like the intended mapping was actually to TRBV19 though, so there seems to be some ambiguity. How would you suggest tidytcells handle this? Should it start emitting warning messages about potential ambiguity whenever it encounters this specific case? Are there plausibly other such ambiguous cases?

Vα7.2

Good shout, I agree. Is there a source I can look at to learn what the mapping rules are between this old convention and the modern one?

[JamieHeather]

Hello hello,

So I actually stumbled on this problem when looking at anti-V gene antibodies, which seem to have largely retained the gene names in use at the time those antibodies were discovered. I just happened to know from the literature that a particular antibody clone (3C10) binds to TRAV1-2 (because that's what people use to enrich for MAIT cells with a canonical invariant alpha chain), however if you look at product pages for that antibody (e.g. this one or this one) you can see that they mostly use the Vα7.2 nomenclature, and sometimes 'TCRAV7S2' or similar (and sometimes some much more bizarre/wrong things too).

That's what lead me to find the gene tables I posted in the first issue. That Arden paper is particularly useful as it gives accessions: so you can look up one of the accessions associated with Vα7.2, e.g. M13732, which is now X04938, and has the following sequence:

>X04938.1 Human mRNA for T-cell receptor alpha-chain HAP12 V(a)7.2
ATGTGGGGAGTTTTCCTTCTTTATGTTTCCATGAAGATGGGAGGCACTACAGGACAAAACATTGACCAGC
CCACTGAGATGACAGCTACGGAAGGTGCCATTGTCCAGATCAACTGCACGTACCAGACATCTGGGTTCAA
CGGGCTGTTCTGGTACCAGCAACATGCTGGCGAAGCACCCACATTTCTGTCTTACAATGTTCTGGATGGT
TTGGAGGAGAAAGATATCCAGAACCCT

... which if you BLAST or analyse however you like you can clearly see it's TRAV1-2. Same goes for TRBV19 = BV17S1A1T or BV17S1: old example accession = M27388, updates to M27388.1:

>M27388.1 Homo sapiens (clone HBVT02) T cell receptor beta chain (TCRB) mRNA, VDJC region, 5'end
ATGAGCAACCAGGTGCTCTGCTGTGTGGTCCTTTGTTTCCTGGGAGCAAACACCGTGGATGGTGGAATCA
CTCAGTCCCCAAAGTACCTGTTCAGAAAGGAAGGACAGAATGTGACCCTGAGTTGTGAACAGAATTTGAA
CCACGATGCCATGTACTGGTACCGACAGGACCCAGGGCAAGGGCTGAGATTGATCTACTACTCACAGATA
GTAAATGACTTTCAGAAAGGAGATATAGCTGAAGGGTACAGCGTCTCTCGGGAGAAGAAGGAATCCTTTC
CTCTCACTGTGACATCGGCCCAAAAGAACCCGACAGCTTTCTATCTCTGTGCCAGTAGTAGGAGGAGAGA
TCTGGGCAATGAGCAGTTCTTCGGGCCAGGGACACGGCTCACCGTGCTAGAGGACCTGAAA

... which again is a clear example of TRBV19. I've not checked them all, but I did a few more when I first found this, and the tables in the original two papers I linked seem accurate. However looking at them it definitely looks like there are other example names in one system that would be ambiguous with older IMGT names (which I'm not sure, but looks like it could have been due to the naming system* being formalised before agreeing on which gene was which perhaps?).

Unfortunately I don't know of a modern or digitized version of those conversion tables that you could easily drop in, but I flagged it up here when I noticed it (as tidytcells was the first thing I tried!), so maybe someone else would know of a suitable resource. It seems unlikely though, as those other naming schemes were supposed to have been deprecated over 20 years ago! (Which I guess is maybe also why some genes appear in HGNC and some don't - maybe some were more established at the time of IMGT formalisation?)

I don't know how much of a problem this will ultimately be, but there's people still publishing with some of these old names, particularly the Greek-character containing ones, which seems to be the convention mostly used in the antibodies. My first instinct was the possibility for users to specify a nomenclature, but then I supposed that anyone using such an old scheme is likely to have stumbled on it by accident, and thus they're probably not going to know what nomenclature they're using - so maybe an ambiguity warning is the best option? I'm uncertain, but either way I wanted to make sure it was on your radar!

[yutanagano]

I see. Thanks for taking the time to flag this up to me and explaining things so thoroughly. I am currently in exam prep season but I am going to make time after that to further improve tidytcells and address these issues. 👍🏼

[JamieHeather]

I've just noticed a potentially more easily scraped resource for (at least some of these) deprecated ambiguous gene names: it looks like the 'Clone names' column of the IMGT gene tables includes some, e.g. TRAV and TRBV. It's a bit odd though, as some of them are clearly gene-level labels being applied to (often non-prototypical) alleles, so I think it might require some manual validation, but at least it should be machine readable!