Evaluate interest of using an algorithm for combining results

[chrstphdm]

Description of feature

Hello.

Because all typing tools have their limitations & advantages (datatype, reads length, read depth, resolution, type of genes typable, etc.), it could be interesting to add some of them as modules activable (txs to DSL2) or not and implementing a combining results algorithm.

SweHAL paper use results from Optitype, HLAScan, HLA-VBSeq and SNP2HLA and combine data using a n−1 concordance rule for class I (four software) and class II (three software) alleles. It must be noted that SNP2HLA is using an imputation model (which is not the type of tool we are trying to use in this workflow, isn't it?).

This paper doing pretty the same with 8 HLA typing tools (OptiType, HLA-HD, PHLAT, seq2HLA, arcasHLA, HLAscan, HLA*LA, and Kourami). The combining algorithm seems much more complex to implement (based on concordance scores computed on 1000G WES data for only the 3 class I genes).

I was not able to find other kind of combining results algorithm but as a first step, n-1 concordance could be a good choice.

Is anyone here already have experience in this kind of problem ?
What do you think of this feature ?

Thank you.

[christopher-mohr]

Hi @chrstphdm,

thanks a lot for the suggestion! I still have to check the publications you shared in detail but in general I think we should definitely aim for something like that with the hlatyping pipeline.

As mentioned earlier on Slack, I would start here with creating new modules for other existing HLA genotyping tools (e.g. I would start by adding one for arcasHLA) and then we can continue from there. In the end, one or multiple techniques for combining results could be added to the pipeline as well.

I would suggest to use this issue as a place to continue with the discussion and then create individual issues for the different tasks.

Best
Christopher