Confounding Factors In Gwas

Question

[int11ap1]

Good morning,

I would like to ask a little question. I have to perform a GWAS. However, analyzing my data, I have found that there are 3 populations among my individuals and a lot of mestizos:

Going on with the analysis, I find a lot of false positives (analyzing a quantitative trait). How to deal with mestizos in GWAS? Is it a confounding factor, right?

Thanks.

Answer 1

[Philipp Bayer]

That depends on the statistical method and the software you want to use - if you like PLINK, you can try to run a linear regression using your 3 populations as covariates, as described here in the manual (scroll down to "Linear and logistic models"). This should give you substantially higher p-values, run the analysis with both covariates and without to see the difference.

---

[int11ap1]

I don't know how to create the covariate files...

[Philipp Bayer]

The covariate file in PLINK looks like the phenotype file, here's a basic example with the covariates "AGE" and "ANCESTRY" and the family ID FAM001 and the individual IDs 1 and 2:

FID IID AGE ANCESTRY
FAM001 1 27 3
FAM001 2 35 1

If you save this as "file.txt" and add the covariates in PLINK using --covar file.txt it will use all of these, to restrict the covariates to for example "age" use --covar-name AGE

Source

[int11ap1]

Are you sure it is a good idea to put the populations as covariates? Think that I have a lot of mestizos. Each of them has a different ancestral composition (0.2% of population 1, 0.5% of population 2, 0.3% of population 3). Their genotype may differ a lot even though having similar percentages for each population. So, I think that the only way to solve this is excluding mestizos.

[Philipp Bayer]

If you think that you have subpopulations inside your mestizos, try using EIGENSTRAT instead to correct the alleles themselves for pop stratificiation, then you don't have to use the covariates.