Summary of questions at the bottom.
I have 7 SNPs genotyped in ~800 samples in a single locus (order of 100kb). I was wondering if this was a a reasonable set for imputation. Not looking to impute the chromosome obviously but just the surrounding region. Naively, it seems like I could get some reasonable calls out of it because there are correlated variants in the area per a quick search on the Broad SNAP tool. However, an analysis with IMPUTE2 and SNPTEST yields no additional information beyond the originally genotyped SNPs.
To look into it I tried looking into the imputation algorithm a bit more. I came across this quote which threw me off:
"Astute readers will point out that imputation is more complex than a simple 1:1 linkage disequilibrium (LD) relationship between an observed SNP and the unobserved variant; imputation is really about haplotypes"
I am not an astute reader because it seems to me if you do have an R2 = 1 proxy for (e.g.) a genotyped snp, then you should able to impute the proxy perfectly. Now if you don't have an R2 = 1 proxy then you have to deal with haplotypes and uncertainty, but my analysis doesn't even yield results for perfectly correlated variants (which would be exactly the same as the genotyped snps admittedly, but moreover I'm concerned with if I'm doing this right).
I plan to open another thread concerning my IMPUTE2 and SNPTEST technicalities depending on this thread, but here I'm looking more for basic understanding.
Summary:
I have 7 SNPs genotyped in ~800 samples in a single locus (order of 100kb). I was wondering if this was a a reasonable set for imputation. Not looking to impute the chromosome obviously but just the surrounding region. Naively, it seems like I could get some reasonable calls out of it because there are correlated variants in the area per a quick search on the Broad SNAP tool. However, an analysis with IMPUTE2 and SNPTEST yields no additional information beyond the originally genotyped SNPs.
To look into it I tried looking into the imputation algorithm a bit more. I came across this quote which threw me off:
"Astute readers will point out that imputation is more complex than a simple 1:1 linkage disequilibrium (LD) relationship between an observed SNP and the unobserved variant; imputation is really about haplotypes"
I am not an astute reader because it seems to me if you do have an R2 = 1 proxy for (e.g.) a genotyped snp, then you should able to impute the proxy perfectly. Now if you don't have an R2 = 1 proxy then you have to deal with haplotypes and uncertainty, but my analysis doesn't even yield results for perfectly correlated variants (which would be exactly the same as the genotyped snps admittedly, but moreover I'm concerned with if I'm doing this right).
I plan to open another thread concerning my IMPUTE2 and SNPTEST technicalities depending on this thread, but here I'm looking more for basic understanding.
Summary:
- Is there a minimum # of tag SNPs needed to impute reliably assuming there are correlated variants in the locus?
- Imputation computes genotype uncertainty based on haplotypes, but is that necessary for perfect proxies?