3
$\begingroup$

I have done a DGE recently and have been looking at the DGE list. One of the genes is HLA-A. However, when I dug deeper I realised there are hundreds of HLA-A genes with unique ENSEMBL number (of course it's HLA...). A few other HLA-A genes also show up on the table but are not significant at all.

Considering the way Salmon deals with multimapping reads, can I assume these expressing genes are different enough that they shouldn't be consider the same? Or should the expression of all HLA-A genes considered as one single gene for the purpose of identifying differentially expressed genes?

I do understand that this may be more of a biological question than bioinformatics, but I also want to see if it is a practice pouring the HLA genes (or other variable genes/loci of similar nature) with different sequences together when doing DGE, and whether the statistics would be valid if I do so?

$\endgroup$

2 Answers 2

0
$\begingroup$

It is bioinformatics IMO because HLA is a massive area of human population genetics and its complex. As a personal opinion I think you need further supporting evidence.

If I understand the scenario correctly,

  1. you are confident the data is correct (see below),
  2. a given HLA-A genotype is significant above other HLA-A and HLA-B genotypes

Thats publishable result, cutting-edge potential, you just need to find the given genotype and represent the association within a given disease association study. Yes your result is absolutely of interest particularly if you consider the wider output of association studies (HLA-B). It is trendy to identify a given HLA to a given association. If this is non-pathogen related they HLA exhibits very strong population skews.

I have put a large caveat below, because:

  • you shouldn't underestimate the complexity of this research field and
  • I don't really understand the specifics of your study.

If you want to ditch or include HLA it would require carefully defined biological aims. It is a specialist area of investigation IMO (below).

HLA diversity and its link to population shifts in infection susceptibility is subject to repeated studies. Different populations show very distinct allelic frequencies. For example, "HLA-B*40 Allele Plays a Role in the Development of Acute Leukemia". Distinct link between African population and malaria .. These just being a very small sample in a hugely active area of investigation. The idea is that specific genotypes have very specific outcomes in human health.

Rationale The single gist is HLA this represents the cellular immune response and is considered essential for infection control, with the notable exception of pathogenic bacteria. Each genotype is considered to be uniquely functional and again is well known to exhibit population skews. It is beyond 'trendy' is huge and established field of investigation.

Caveat If you were looking for 'wiggle room' to avoid HLA-A diversity and allelic frequency and all the associated work it requires ... or you specifically want to make it a standout result ... then the angles that might work are ...

  1. alot of examples of association focus more on HLA-B than HLA-A. This is a rule of thumb and might be disputed. So if you're going to ignore HLA-A you could perform an extensive literature review to see the extent that associations of HLA-B predominate. I think you might find that holds. So its either really cool cause its an outlier, or you can ditch it because its an outlier.
  2. If pathogenic bacteria were the focus of your investigation, then most would accept HLA is not relevant.

I'm sure there's other angles of justification, e.g. "I recognise HLA-A .... but this will be subject to further investigation by (putative) collaboration".

Advice If I was tackling this data I would contact a bioinformatics immunologist, because they will have databases and pipelines dedicated to the exact task you require. Personally, I would be reluctant to take this on without specialist assistance, at a minimum someone with current experience. Essentially, what I'm trying to say, is you will need to understand the HLA field because crossing into it isn't trivial. The situations where I could be wrong are that its simply so complex that no-one really understands it, in which case your good to go.

$\endgroup$
4
  • 1
    $\begingroup$ That's really well explained! Thanks @M__ ! So the research question I have is basically the transcriptomic difference between haematological malignancy which enter remission and those which relapse. This may sound stupid but are these unique ENSEMBL IDs of HLA-A just a different allele/genotype/serotype (I am not familiar with the terminology)? And since each person has only a pair I can tell which alleles they have by RNA-Seq (theoretically)? $\endgroup$
    – Kento
    Sep 1 at 11:44
  • 1
    $\begingroup$ I don't know this data set. The data should be individual HLA-A genotypes if correctly annotated, which they may not have done. The second part of the question is definitely yes you can tell if correctly annotated (I don't understand 'one pair'). This may help raw.githubusercontent.com/ANHIG/IMGTHLA/Latest/alignments/… hla.alleles.org/data/hla-a.html Finally, there are no serotypes in HLA, for an explanation of terminology ncbi.nlm.nih.gov/pmc/articles/PMC3715123 . $\endgroup$
    – M__
    Sep 1 at 12:28
  • 1
    $\begingroup$ Thanks a lot! Just to add in case people are reading, the unique ENSEMBL IDs are from MHC contigs from different cell lines. I will try to see if it is possible to get the genotype of the samples by WGS and align RNA-Seq reads accordingly. To clarify, when I say pair I assume like other genes in the genome there are two alleles hence a pair of the genes. I will look up more about HLAs and MHCs. $\endgroup$
    – Kento
    Sep 1 at 15:16
  • $\begingroup$ @Kento MHC is basically the same thing, so that data looks fine. However, if this is a mouse model then (obviously) they don't have HLA just MHC. $\endgroup$
    – M__
    Sep 1 at 20:10
3
$\begingroup$

You need specialized tools to study HLA genes. Generic tools won't work well. Before other analysis, you should get the correct genotype first. Try arcasHLA. It has been a popular HLA genotyper for RNA-seq data. When you know the genotype, I guess you may get rid of other HLA-A alleles and redo the differential analysis if you are interested in HLA-A. I haven't done this, though. Just a thought.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.