Correcting for population substructure in a small sample

Question

[kylec1729]

I'm running several case-control association analyses on ~50 SNPs with around 1000 cases and 1000 controls. My only "correction" for population stratification so far has been to select only Caucasian British participants, but does this really account for population structure? Are there other things I can and should do to account for population structure? If it matters, these SNPs are all on the same chromosomal locus.

Answer 1

[Asaf]

I guess you can't do a lot with what you have. If all the data you have is from the same locus you have no way of evaluating the population structure without loosing the effect.

[kylec1729]

That's too bad, I guess I kind of figured that. If I decided to do a GWAS or a chromosome-wide AS, is PCA the best method for correcting for this?

[Kevin Blighe]

Indeed, PC eigenvectors as covariates, but only those eigenvectors that actually stratify the population(s) via bi-plots, or that have statistically significant p-values when regressed to population. I notice that a few groups naively include XY number of eigenvectors without even checking whether or not they are related to population structure.

[kylec1729]

Thanks Kevin. Do you have any references that I can read on how to stratify via bi-plots/regress to population?

[Kevin Blighe]

It's such a common procedure such that it will be mentioned in many published manuscripts. For example, I just went to my search engine and searched for ncbi control population pc eigenvectors, and many hits came up.