Clarity regarding indep pairwise function

1,346 views
Skip to first unread message

Palvi Malik

unread,
Jun 22, 2018, 10:29:00 AM6/22/18
to plink2-users
Hello 

I am not able to understand indep-pairwise function. Supposedly, I use the following command:
plink --chr-set 12 --vcf filename --indep-pairwise 50 10 0.1

I understand that it would consider SNP window of 50 SNPs, compute pairwise correlation among them and if value > 0.1, it would keep one SNP out of the pair. 

In the first window of 50 SNPs, it would compute correlation btw 1 and 2, which one SNP would it delete if correlation > 0.1? 

If correlation between SNP 1 and 20 < 0.1, it would retain both of them. For making next comparison, will it work out pairwise correlation between SNP1 and 21 as well between SNP 20 and 21? I want to know if this function calculates r2 between pruned SNPs (1 and 20, in this case) and remaining SNPs?

Let's assume there are only 3 SNPs left in the first window of 50 SNPs. I don't understand will it consider next 47 SNPs to perform the analysis in a similar way? But if this the situation, then I don't understand the relevance of second parameter (10 in this case)?

Does this function calculate r2 among different chromosomes?

Can someone help me understand this?

Cheers
Palvi Malik

Christopher Chang

unread,
Jun 22, 2018, 12:47:53 PM6/22/18
to plink2-users
- When two variants in the same window are too highly correlated, PLINK removes the variant with the lower MAF.  (If you want to prioritize in a different manner, you can 'trick' PLINK into doing so by feeding it fake MAFs with --read-freq.)
- When the second parameter is 10, PLINK will start by checking pairs of variants in {#0, #1, ..., #49}, then it'll check pairs of *remaining* variants in {#10, #11, #12, ..., #59}, then it'll check pairs of remaining variants in {#20, #21, #22, ..., #69}, etc., up to the end of the chromosome.  If all of the variants in the first window are highly correlated with each other and #12 had the highest MAF, then the second window will start with only 11 variants: {#12, #50, #51, ..., #59}.  And #12 may be removed if e.g. it's correlated with #50 and the latter has a higher MAF.
- The LD pruning functions never compare variants between different chromosomes.
Reply all
Reply to author
Forward
0 new messages