GxE analysis with LRT

[saudat-bio-code]

Dear flexlmm team,

I have a quick question regarding model structures for GxE analysis.

I am hoping to extract p-values specifically for GxE component. Ideally I wanted to have LRT test with GxE as the term that is distinct between two models, while having G + E in the reduced null model. Thus my code would look like this:

--model_formula 'y ~ x + x:environment + environment'
--null_model_formula 'y ~ x + environment'

However, the analysis stops at the FIT_NULL_MODEL step. My formulas are valid and pass VALIDATE_FORMULAS without any issues.

Of course, the manual mentions that x should not be included as the term in the null_model_formula. However, when the focus of the analysis is solely GxE, it makes sense to have G in the null model. Do you have any recommendations for solving this issue?

This is the line from the manial that confirms it: NOTE: the null model formula cannot contain the term x or functions of it. The covar and qcovar file cannot contain columns named x or y.

My other currently is to run two models

--model_formula 'y ~ x + x:environment + environment'
--null_model_formula 'y ~ environment'

and

--model_formula 'y ~ x + environment'
--null_model_formula 'y ~ environment'

and only then manually use LRT to get p-values that describe GxE. I think more intuitive way would be to have gxe test as described above, but I would need some advice in stopping FIT_NULL_MODEL from failing. Currently it quits with exit status 1 and no output.

Thank you for your time and help!

Saudat

[saulpierotti]

Dear Saudat,
for performance reasons it is more efficient to not to allow for x in the null model, since the null model is fit only once at the beginning and not for each SNP. Your workaround to manually run the LRT I think is a good approach!

In the next release of flexlmm I can add support for x in the null and fit the null snp-wise when x is included.

[saudat-bio-code]

Dear Saul,

Thanks for the clarification! Yes, it would be really helpful to have x in the null model for proper GxE tests.
Looking forward to the next release of flexlmm! Thanks!