question about PCA in plink 1.9

Question

[summeryhx]

I'm new in biostatistics and I work on something about GWAS with plink.

I used plink 1.9 to do the PCA with the code as following:

plink --bfile data --pca

and get the .eigenval and .eigenvec file.

I think the results of eigenval file (as following)is wired, is that mean the amount of variance accounted for by the principle component? if it is, should I count the percentage variance explained by certain PC divided by total variance? But the percentage is too small.how many eigenvectors should I choose then?

20.0134
2.98845
2.32333
1.94295
1.93421
1.91117
1.88628
1.86544
1.85781
1.84763
1.76204
1.5532
1.3277
1.1808
1.14857
1.13482
1.13316
1.12439
1.1194
1.11312

Answer 1

[chrchang523]

The first eigenvalue is far larger than the rest, so I don't understand what you mean by "the percentage is too small".

Given this spread of eigenvalues, it would make sense to use 1-3 PCs as covariates.

[summeryhx]

Thanks for your help. But we should choose the top m pcs which percentage add to at least 85%, is that right? And my eigenval file is just the variance, not the percentage, when I get the percentage,it will be too small

[chrchang523]

That depends on what exactly you're trying to do with the PCs; I don't know what scenario the 85% rule you mention was intended for. But your "too small" comment still does not make sense. Go ahead and take the top 10-15 PCs instead of the top 3 if that's appropriate for the analysis you are performing; but there's nothing wrong with the output file itself.