Hello, I have some theoretical questions.
I am performing genome-wide association (GWAS) mapping using GEMMA. I have 150 individuals and 883207 SNPs. I am using the lmm
model with a kinship matrix as a random factor. The p-values are estimated using wald statistics. In this way, I see the QQ-plot is inflated. Then I estimated the genomic inflation factor which is 0.98. Does that mean the confounding effect from population stratification is not affecting the GWAS?
Then, I added 3 principal components as covariates along with kinship in the GWAS. In this way, I see the QQ-plot has improved and the genomic inflation factor after adding the PCs is 0.94. For some traits, adding the PCs does not change the genomic inflation factor at all.
However, as I have different traits, adding PCs+kinship in the GWAS model leads to the removal of significant associations that I obtained only using LMM+Kinship. Also in some cases, adding PCs leads the estimation of percent variance explained (PVE) by GEMMA close to ZERO.
Can someone please shed some light on these issues? Should I continue to include PC in the LMM+Kinship model? or should I add the PCs to the GWAS model for some traits and do not include it in other traits?
How do you explain the reduction in the genomic inflation rate after adding the PC?
Looking forward to the feedback.
Thank you.