I am using regression models to test the effect of two genetic mutations (let's say SNP1 and SNP2) on a cognitive error-evaluation measure (let's say CEEM), and I am controlling for age. I run the following models:

CEEM ~ SNP1 + Age

CEEM ~ SNP2 + Age

CEEM ~ SNP1 * Age

CEEM ~ SNP2 * Age

Here, there are significant main effects (F) of each SNP on CEEM as well as significant interactions between each SNP and Age on CEEM. The regression models are showing 'significant' effects.

Because my two genes contribute to a similar molecular mechanism, I combine them by a simple summation method, unweighted genetic risk score (uGRS), and run the models:

CEEM ~ uGRS + Age

CEEM ~ uGRS * Age

Here, the same story - significant main effect of uGRS and significant interaction.

In my first, single-gene models which include '+ Age', the slope (beta) is nonsignificant. So, my question is: complex interactions aside, to what extent does that matter? Can I report these beta-coefficients, their associated t-tests and significance levels without concern?


I think the specific answer to your question is yes, you can report the results of your tests, but with concern. It appears that you are performing an association genetic test with linear regression. There are several studies in the last ten years that demonstrate the weakness of linear regression. I suggest reading Yang et al. 2014 and using a mixed effects model. I assume it's important for your phenotype to correct for age, but it may also be important to correct for genetic structure from your samples. Every GWA model I've come across has a term to control for population structure. Your SNPs of interest may have the same allele frequency as countless other neutral alleles in your population - you need to account for this.

Also, how do you create the uGRS variable? Have you considered performing an ordination of the SNPs, and then taking PC1 as input into your regression? I can't find the source I'm thinking of that uses this method, but its from the human GWA literature.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.