For a little context here: I have been working on a project which has involved a large-scale GWAS meta-analysis using yearly healthcare cost as a phenotype. One of the major signals for every phenotype we have used is in the HLA region and so we are trying to disentangle this a little, trying to get a clearer picture of the source of the association.
The first step was to conduct a straightforward HLA association test of our phenotype. We have access to both the FinnGen and UK Biobank cohorts and both of these provide HLA imputation/genotype panels in tabular form. E.g. something like the following for each individual (where the nomenclature means that A_101 corresponds to the HLA-A*1:01 allele):
A_101, A_102, A_201, A_301
0, 1, 0, 2
0, 2, 1, 1
0, 0, 1, 1
We did a very straightforward assocation test of phenotype ~ hla_genotype + covariates (for each imputed genotype) and then meta-analysed those which overlapped, and got a bunch of significant hits.
However, we don't think this is quite enough and in particular we need to do some form of conditional testing so that we know which are the actually important alleles within each HLA gene/protein. The problem is that from here on, my expertise ends and I am pretty stumped with how to progress.
I have been looking at this paper: https://pmc.ncbi.nlm.nih.gov/articles/PMC10786448/ which mentions conducting something they call an Omnibus test, but to be totally frank I simply don't understand exactly what it means or how to conduct it. It seems to involve amino acid permutations downloaded from somewhere on https://www.ebi.ac.uk/ipd/imgt/hla/release/ but this really isn't super clear to me how to actually incorporate it.
Another option from https://academic.oup.com/bioinformatics/article/37/3/416/5879278 provides a tool called HATK for basically doing the whole analysis (and according to their GitHub this actually includes an OmniBus test) but has the issue that the data seems like it needs to be in a particular format, and I am actually struggling to understand what format that is (the GitHub makes repeated reference to a Wiki I couldn't find which doesn't help).
Essentially, I am wondering if anyone can provide any guidance on next steps for our analysis here. If they can provide some kind of guidance for doing an Omnibus test, using HATK or even some other, simple workflow?
Any input greatly appreciated!
