GWIS configuration

[joshua-slaughter]

Hi all, I have put together a configuration for estimating pairwise interactions at the genome-wide scale. The changes to make this possible only required minimal changes to inputs_from_config.jl. A review would be very appreciated!

[olivierlabayle]

Thank you Josh, I'll have a look asap!

[olivierlabayle]

Thanks Josh, see my comments in the review, happy to discuss.

[olivierlabayle]

@joshua-slaughter Could you provide a high level overview of the new functionality here and how a user is meant to interact with it? This description could also be used in the next release of the TarGene docs.

[joshua-slaughter]

@olivierlabayle had to dust this one off haha

Notable changes

• In the GWAS configuration the keyword extra_treatments can be specified to add treatment variables to be used for higher-order interactions. I have recently updated this to go beyond environmental variables but to also include genetic variants as well enabling genome wide epistasis analyses. However, this could probably be cleaner (e.g. right now it is more similar to the flat config rather than groups)
• Added output for the pipeline to track genotype counts as this has been of particular interest to the group (questions about case control analyses)
• Tests have been added to evaluate that the new functionalities behave as expected.
• Upped the CI.yml

[joshua-slaughter]

@olivierlabayle had to dust this one off haha

Notable changes

* In the GWAS configuration the keyword `extra_treatments` can be specified to add treatment variables to be used for higher-order interactions. I have recently updated this to go beyond environmental variables but to also include genetic variants as well enabling genome wide epistasis analyses. However, this could probably be cleaner (e.g. right now it is more similar to the flat config rather than groups)

* Added output for the pipeline to track genotype counts as this has been of particular interest to the group (questions about case control analyses)

* Tests have been added to evaluate that the new functionalities behave as expected.

* Upped the CI.yml

Will work on docs for this soon but would like to get your opinion on the current implementation if possible (at least for the interaction setup)

[olivierlabayle]

Hi Josh, thanks for the PR, please find my comments inline. Main things are:

1. I think the summary stats (counts) should either be moved to TMLECLI as an option or kept here but in a different function. It would also be good to discuss exactly what we want to countI suggest to make use of DataFrames.jl groupby combine in any case to simplify the code and make it faster.

2. The GWIS looks mostly good while I think it is getting quite complicated and requires a bit more testing of edge cases that are not covered.

Thanks!

[joshua-slaughter]

@olivierlabayle addressed all comments. Will open summary stats PR in TMLECLI.