As many SNPs are in LD, my colleagues and I usually test all of them, even impute the missing SNPs to find the optimum, and then the correction for multiple testing shouldn't be Bonferroni but less since we are not testing as many independent hypotheses as SNPs. That is how most GWAS etc. go.

A colleague and an editor stated to us that is such cases, it is not legitimate to do individual SNP analyses since each of these 9 SNPs tag the other 8 SNPs. The correct approach is to perform haplotype analyses which will yield two yin yang haplotypes of nearly equal frequency from which three diplotypes will be derived.

I disagree. My own opinion is that deriving haplotypes will create more errors than running each SNP separately. when there is complete LD, it shouldn't matter, except for technical artifacts (one SNP not being genotyped quite correctly), so my sense is that it is better to pick the best SNP (as in: most completely genotyped without being out of HWE), and use that for association, not to create haplotypes.

Similar questions and discussions