|
Post by wei on Dec 4, 2014 21:56:38 GMT
Here we have a quantitative trait. So I guess low maf should not be a QC problem. I see the benefit of including low maf snps, which are supposed to capture rare causal ones better. But on the other side, the formula for Aijk has p*(1 - p) in the denominator (following the nat. gen. 2010 notation). A p close to 0 would generate a huge and unstable value. So what's the recommended approach? The MS paper (from nat. 2012) seems to just use the entire frequency spectrum. thanks wei
|
|
|
Post by guest on Dec 6, 2014 12:29:43 GMT
In GCTA, the genotypes have been standardised, z = x/(2*p*(1-p)), where p - allele frequency, x - 0, 1, 2 for three genotypes, z - standardised genotypes, which supposes the contribution of each SNP is equal (rare SNPs doesn't contribute to SNP heritability too much). If you are using GCTA to estimate SNP heritability, I would suggest removing SNPs with low MAF.
|
|