GWAS-by-subtraction sumstats genetic correlation

[Marta Ribases]

Hi,

I run GWAS-by-subtraction and apparently it worked well. However, what I found unexpected is that the genetic correlation between the resulting sumstats was strongly negative (rg=-0.38, P=1.0e-31). Given that the covariance between both latent factors was set at 0, I just wonder if this could be an artifact of the method.

Many thanks in advance,

Marta

[Lars Andreas Rømo Ystaas]

I have the same problem. I’ve run a GWAS-by-subtraction successfully, but there is still a residual correlation in the factors as estimated by LDSC (rg=0.1488, p=0.0003). I’m running GWAS-by-subtraction on multivariable GWAS factors from a previous GenomicSEM run, so the number of SNPs is quite low (about 2.1 million), could this introduce some error when merging with the reference panel perhaps?

Here’s the formula I used:

F=~NA*X1 + start(0.35)*X1 + start(0.12)*X2
NonF=~NA*start(0.08)*X2

F~~1*F
NonF~~1*NonF
NonF~~0*F

X1~~0*X1
X2~~0*X2
X1~~0*X2

F~SNP
NonF~SNP
SNP~~SNP

Appreciate any insight anyone might have!