flashPCA2 v PLINK2 pca approx

[Vince Forgetta]

Hi,

Quick question.  Should flashPCA2 give near identical results compared to PLINK --pca approx? In the past I have observed identical results when using EIGENSTRAT.

I have run flashPCA2 on a data set pruned/cleaned like so:

     plink --bfile ../bed/${i} --keep-allele-order --maf 0.0001 --geno 0.05 --hwe 0.000001 --remove IDsToExclude.txt --exclude $SHARE/DATA/GWAS/QC/high-LD-regions-hg19.txt --range --indep-pairwise 50 5 0.8 --out $i

Note the high LD cutoff. This was intentional as my aim is to use this snp set as input into BOLT-LMM for computing the GRM, and wanted to include as many SNPs as possible to capture relatedness within UKB.

I ran flashPCA2 like so:

    flashpca --bfile eur --suffix .flashpca

I ran PLINK --pca like so:

    plink2 --bfile eur --pca 20 approx --threads 20

I only realized after that flashPCA2 will return first 10 PC, where as PLINK returned 20 PC.

The first and second PC are highly correlated, but not perfectly so.  I assume his may be due to the number of PC returned being different, but may also be differences in the algorithm.

Your thoughts on this are most appreciated.

Thanks,

Vince

[Vince Forgetta]

I also generated 10 PC using plink:

    plink2 --bfile eur --pca 10 approx --threads 20

Same issue in that PC1 is not perfectly correlated between plink and flashPCA2. Correlation ~ 0.993.

Thanks,

Vince

[Gad Abraham]

Hi Vince,

Did you solve this issue?

Gad