Why is it often necessary to perform z-transformation on quantitative data when doing GWAS?

Question

[Isaac]

Hello everyone. I would like to ask a question. I have found that in mainstream quantitative phenotypic GWAS, phenotypic data is converted by INT or log first and then use z-transformation. I don't quite understand, what is the necessity of z transformation in the second normal transformation strategy? Doesn't this lead to poorer interpretability of beta values at the site?

[Asaf]

I think it stems from the belief that z transformation magically transforms your data to normally distributed. I agree that it makes it harder to interpreter the beta values and I can't see a benefit in the logistic regression.

[Isaac]

Thanks for answering! In my opinion, I don't think the z-transformation will be very helpful in driving normality. In fact, it hardly changes the shape of the histogram, but many people do it, and I'm not quite sure if my idea is correct.

[Asaf]

It doesn't change the shape, only the mean and std. In some cases when you want to remove a confounding factor before you run gwas, you can run a logistic regression and take the residuals for the gwas, in this case they will be z normalized.

[Isaac]

Thanks! Actually, I have another concern. I believe that if there are certain differences in the distribution of two populations, converting them separately and performing meta-analysis on the two GWAS results will lead to heterogeneity between the results.