KING --unrelated --degree 2 vs PLINK --king-cutoff 0.125

[mizetrav]

Hello,

I am attempting to remove 1st and 2nd degree related individuals from our trans-ancestry sample. I am getting different results when I run KING vs PLINK's implementation of KING. KING is providing a list containing 1119 more individuals to be removed than PLINK (log files attached).

Here are the commands being run:

plink2 --bfile ${bfile} --king-cutoff 0.125 --out ${out}

king -b ${bfile}.bed --unrelated --degree 2 --cpus 10 --prefix ${out}

Here are the versions:

PLINK v2.00a2.3LM 64-bit Intel (24 Jan 2020)

KING 2.3.2 - (c) 2010-2023 Wei-Min Chen

Here is the QC I have performed up until this point. ${keep} is a list of SNPs from the Global Screening Array that hopefully captures high quality SNPs in our highly diverse sample:

plink2 --bgen ${bgen} --sample ${sample} --extract range ${keep} \

--geno 0.05 --maf 0.05 --max-alleles 2 \

--set-missing-var-ids @:#\$r:\$a --new-id-max-allele-len 1000 \

--make-bed --out ${bfile}

Any advice/guidance you can provide would be greatly appreciated!

Thank you,

-Travis Mize

[Chris Chang]

The standard KING cutoff for removing 1st and 2nd-degree relations is ~0.0884, not 0.125.  0.125 is the *expected* kinship coefficient between 2nd-degree relations; plenty of 2nd-degree relations will have slightly-lower kinship coefficients due to random chance.

--
You received this message because you are subscribed to the Google Groups "plink2-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to plink2-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/plink2-users/bb9a1080-de74-4ecc-b964-0b9b76eed70bn%40googlegroups.com.

[mizetrav]

Hi Chris,

Thank you for the very speedy reply!

I apparently misunderstood these values, thank you for that clarification it is greatly appreciated. When I rerun using 0.0884 as the cutoff, PLINK now provides a list of 4217 individuals to be removed. This is still a discrepancy of 379 individuals. I assume there will always be some difference as the algorithms used to determine the final sample is not the same between KING and PLINK (if I am understanding correctly), but should it be this large?

Thanks again,

-Travis Mize

[Chris Chang]

Although the algorithm used by plink2 --king-cutoff to select unrelated samples is rather simple (see e.g. https://groups.google.com/g/plink2-users/c/gb6oh2eCwI0/m/lfeIHA2AAAAJ ), it consistently manages to keep more samples than other pruning algorithms I've seen in the wild.  379 more samples on top of 53788 is not surprising to me.

You should be able to e.g. use KING to confirm that the plink2-generated list does not contain any pairs of 1st- or 2nd-degree related individuals.

To view this discussion visit https://groups.google.com/d/msgid/plink2-users/4b2ce56f-d873-40a6-a07d-d6dbcf8dc437n%40googlegroups.com.

[mizetrav]

Great thank you so much for the discussion, Chris.

Take care,

-Travis Mize