|
Post by xediii on Sept 22, 2015 9:38:38 GMT
A way to correct for population stratification is to include PCA as covariate in GREML analyses. My understanding is that the variance explained by the GRM may be biased, if ancestry is correlated with environmental effects. Thus I expected that inclusion of PCA as fixed effects in the GREML model will decrease the SNP heritability estimate in presence of population stratification, or have no effect. However, I observed an increase from 0.17 to 0.28 for one phenotype.
How is this possible? The PCA are based on the genotype, so its effects are included in the GRM, when the analysis are not adjusted. How can then the GRM effect increase, if one corrects for one component of the GRM? Does the total phenotype variance change? Since the heritability is Vg/p, is p now the phenotype variance without variance explained by ancestry?
|
|
|
Post by Jian Yang on Sept 24, 2015 22:10:25 GMT
If the sample size is small and SE is large, this could just be due to sampling.
|
|