Population stratification with PCA

Question

[Yong]

Hi all! I have a genotype dataset in plink format. Now I want to correct for population structure with PCA in association analysis. I split my dataset to training and testing datasets. I want to do the PCA only in the training dataset and use the training dataset as a reference panel to calculate the PCs in testing datasets (I don't want to combine them before PCA). Dose anyone know how to do that? Thank you!

Answer 1

[lassefolkersen]

Yes, you can multiply the genotypes of the test data (as alt-allele counts) with those from eigen-vector from the training-set-only PCA. That's very fast to do. I use it to derive per-person placements in a PCA plot of 1kgenomes individuals, like this for inspiration, and it takes literally just a split-second to get.

I've used the prcomp package in R for that calculation, you could follow the calculation in above code-link if you want (I think the original pca-call is here), but the principle is the same for any PCA-package.

Good luck!