Hi

If I was true, you are trying to run a GWA analysis. in such cases, you should use principle component analysis (PCA) to consider stratification and use some PC level as co-variate in your model (based on the proportion of variation). Or if you have a relationship matrix - Genomic or pedigree-based - (such as is common in animal population) you can use a Mixed Model with this matrix.

wish be helpful

Good luck

Hi Mehdi,

Thanks for the reply. You are right, I am doing association tests and I am already using PCA (EIGENSTRAT methods) to correct for stratification. I wanted to compare the GC (Genomic Control) approach with EIGENSTRAT and came across this issue. You are right, LMMs can also be done in order to achieve the same. I was doing GC for verification purpose that on this simulated data set, indeed PCA based approaches work better than GC as is in the Price 2006 Nat. Genet. paper.

Thanks again.

Your welcome Aritra

some tips that you may know:

1- you shouldn't use GC when lambda inflated more than 10% (lambda > 1.1), because GC doesn't consider stratification. It just reduces lambda.

2- EIGENSTRAT method adjust both phenotypes and genotypes for possible stratification by using PC levels. this is a good method, but I recommend you use some steps before EIGENSTRAT, such as using just principal components of genetic variation in the model (genotypes PC levels). It worked well for me.

3- In some cases phenotypes PC levels can be used, too. It's not more common.

Good luck

Thank you Mehdi. The first point particularly is very insightful. So that's what's going on I think!