PLINK GRM versus GCTA GRM

[Daan van Beek]

Hi all,

I ran --make-rel and --make-grm-bin both with and without the "cov" modifier as well as a --make-grm in GCTA on the same, qc'd bfiles (maf > 0.05, hwe < 1e-6, call rate > 0.95). I would expect, since PLINK refers to the GCTA paper that the --grm results would at least be the same. However, to my surprise the PLINK results from --make-rel and --make-grm were the same in their values and (--cov just divides all values by 2), but the --make-grm from GCTA is different to the GRM from PLINK but both should be processed the same into a binary file and I loaded them the same in R.

Does anyone know and want to explain to me where the difference in GRM values comes from? :)

Best and thanks in advance,

Daan

[Dominick A. Leone]

Hi Daan,

I have not used either PLINK or GCTA to create a GRM, but when I use pcrelate (from the GENESIS R package), you can change the scaling of the kinship coefficients (nxn matrix). Ignoring the methodological differences between GENESIS and PLINK or GCTA, I would start by examining any options for PLINK and GCTA. Also, check the diagonals: are they the same between PLINK versus GCTA? For example, using the GENESIS approach, the GRM diagonals are scaled by 2 (default) when converting a pcrelate object to a matrix (GRM); this is similar to —cov in PLINK. Furthermore, some software (like GENESIS/pcrelate) use PCs when calculating the GRM, and there are methodological differences (options in the software) that might explain why the matrix coefficients differ across different software packages. 

QUESTION: if you compare associational results from PLINK and GCTA for the first 20 SNPs (on say chr22), are the betas and p-values similar? The betas should be almost identical, and the p-values should be “close”. 

Best,

Dominick Leone, MPH, MS
Doctoral Candidate, Epidemiology Department

Chronic Kidney Disease in Central America Research Group    
Boston University School of Public Health

801 Massachusetts Avenue

Biostatistics Dept; Suite 345K
Boston, MA 02118
 
Phone: (617) 893-9493

Email: nle...@bu.edu

 

THINK. TEACH. DO.
FOR THE HEALTH OF ALL.

--
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 on the web visit https://groups.google.com/d/msgid/plink2-users/74ef7add-8409-4e6e-8ba1-d41554f4378dn%40googlegroups.com.

[Christopher Chang]

My first guess would be a difference in the minor allele frequencies.  PLINK usually excludes "nonfounders" (samples which have at least one parent listed in the .fam file) when computing allele frequencies, and I don't think GCTA does; do you have any such samples?