Post by Jian Yang on Sept 30, 2015 11:29:44 GMT
Q: Can I run a GREML analysis using a subset of SNPs selected by p-values from GWAS?
A: If the SNPs are ascertained by p-value from GWAS analysis in the same sample, the GREML estimate of variance explained by this subset of SNPs will be inflated due to the winners' curse issue, i.e. the selection creates a positive correlation between true SNP effects and estimation errors.
If the SNPs are selected by p-values from association analysis in an independent sample, then it's OK. For example, in Wood et al. 2014 Nat Genet, we selected SNPs in a discovery set and performed GREML analysis of the selected SNPs in an independent validation set.
A: If the SNPs are ascertained by p-value from GWAS analysis in the same sample, the GREML estimate of variance explained by this subset of SNPs will be inflated due to the winners' curse issue, i.e. the selection creates a positive correlation between true SNP effects and estimation errors.
If the SNPs are selected by p-values from association analysis in an independent sample, then it's OK. For example, in Wood et al. 2014 Nat Genet, we selected SNPs in a discovery set and performed GREML analysis of the selected SNPs in an independent validation set.