GWAS on dosage data - interaction

Question

[Abdel]

Is there a software package for GWASs that can include an interaction effect (SNP x predictor) in the regression using dosage data?

Answer 1

[coleman_jonathan]

Can you use PLINK2?

That provides an interaction term for the linear regression... In terms of the dosage data, do you have to use the probabilistic data, or would best-guess hard-calling work?

[Abdel]

My best-guess genotypes unfortunately don't work as well as my dosage data for some reason...

I'll check if plink2 can handle dosage data in this way, thanks!

[chrchang523]

PLINK 2.0 will have proper support for dosage data, but 1.9 cannot since it is limited by the PLINK 1 file format.

SNPTEST may be your best bet for this: https://mathgen.stats.ox.ac.uk/genetics_software/snptest/snptest.html

[Tafelplankje]

snptest does not handle SNP interactions.

[Tafelplankje]

that version does not handle dosage data in combination with interaction...

Answer 2

[Devon Ryan]

Assuming you're using a logistic regression for the underlying statistic, then you could do that in R easily enough. I'd also be very surprised if plink can't do that too.

[Abdel]

It's a linear regression. And Plink unfortunately can't do that...

[Devon Ryan]

I'm surprised that plink won't handle that, that would seem to be a useful feature. Either way, you can just do it in R, which will handle pretty much any model you want.

[Tafelplankje]

Large datasets and R are a pain in the ass...

[Devon Ryan]

Depends on how much memory you have, but yes, it can become problematic.

Answer 3

[joeyinjie]

I have the same problem recently. Just want to ask if PLINK2.0 handled your problem that - - interaction doesn’t run with dosage data. Thanks!

[chrchang523]

plink 2.0 —glm handles dosage data and interactions.

[joeyinjie]

Thanks a lot! I’ll try it.

Answer 4

[Tafelplankje]

you could also use quicktest http://toby.freeshell.org/software/quicktest.shtml