williamdlees
commented
2 years ago Thanks Scott. This is an important question and I'll do my best to address it succinctly.
Leaving tooling workflow aside for the time being, and referring to the germline documentation , a germline set is a curated set. Normally it would exist in a repository, and normally it would represent all the genes and alleles found in a species or a sub-population. A genotype is the subset of genes and alleles found or inferred in an individual - listed with reference to a germline set, but allowing for additions and exclusions. For an analogy, think of a questionnaire in which the answers to each question are multiple choice, but with the option for the responder to provide their own free text answer, or to decline to answer a particular question altogether. If each question is a gene and the multiple choice answers are the alleles, the set of all questions and multiple choice answers is like a germline set. A specific response to the questionnaire is like a genotype. You could envisage storing the complete questionnaire alongside each response, as both are needed to understand the response in full, but it would probably be better to store a single copy of the questionnaire somewhere, and refer to it when storing a response, in case there were multiple versions of the questionnaire.
This is what we do in VDJbase. If you go to a samples page, and click on one of the icons in the Genotype column, you'll see a bunch of reports which detail the genotype and some statistics around V-gene usage. They don't include a germline set. A reference to the germline set with which the samples were annotated is available in the sample metadata (click on the name of a sample in the first column to see this). This is what I would recommend for an ADC - store a reference to the germline set, which can be stored centrally in a location managed by the ADC, or in a trusted third party repository such as OGRDB. Store a genotype with the repertoire, and base statistics on gene usage etc on a repertoire which reflects the genotype.
Turning to tools and how they work - I don't think this is so relevant to the question but as you brought it up it may be worth covering. Tigger is a function library, which can be used in various ways, but the core use case doesn't involve creating a modified germline set. It has some functions to infer a genotype, in its own format (you can see the Tigger format in VDJbase, under that genotype icon). It has another function which will add the genotyped V-calls to a changeo file. This doesn't involve a re-run of the annotation tool. IgDiscover has a rather different use case and typically runs multiple passes of inference and annotation, for which it does use modified germline sets. Modified germline sets may therefore be used by the tooling workflow, but they are not mandated. In VDJbase and OGRDB we use a tool called ogrdbstats which, among other things, derives a genotype in a single format from the output of the four inference tools we know about: Tigger, IgDiscover, partis and Impre. It doesn't derive a tailored germline set.
Why might it be a bad idea to store tailored germline sets with each repertoire in the ADC? Firstly there are space and complexity issues, the latter of which you alluded to in the issue. Beyond that, as I see it, one function of the ADCs and their gateways is to provide usage statistics for particular alleles, clonal information and so on. These things depend for accuracy and consistency on the underlying germline set used for annotation. Since the IARC was established (roughly the same lifetime as the ADCs) we've affirmed about 30 novel human alleles, and I would say that IMGT has approved at least as many again from genomic sources. Today we know of maybe 200 more from sources available now but not yet confirmed and published, across the human IG/TR loci. So even in the human, which is well studied compared to any other organism, the germline set is not stable now, and will not be stable for some time. If I search for a recently affirmed allele in the ADC, how many repertoires will have been sequenced with a germline set that defines that allele? Likewise if I look for clonal families that are based on that allele? I think it's important to know the base germline set on which annotation was conducted - much more important than having access to a personalised set, which, if needed, can always be reconstructed from the base set and the genotype. And important also, I would say, to re-annotate reasonably frequently with an up-to-date set, in order to maintain consistency of analysis across the repertoire collection. OK, this will be computationally expensive - but maybe dedicate say 5% of available capacity to re-annotation, or whatever can be afforded, and see how far it takes things.
schristley
bcorrie
If we get to point of using subject-specific germline sets, this has the potential to create many germline sets, potentially one for each subject repertoire in the ADC. Is this the desired design?
By subject-specific germline set I mean a workflow like this:
When that repertoire is loaded into the ADC, so too does the custom germline set.
I think this is the most obvious design but there are alternatives. This might also inform the decision about normal/denormal for
AlleleDescriptioninGermlineSet. Presumably in the above, there would be many germline sets but the allele descriptions would be the same record across many of them.