Lets say I am running a GWAS for a disease condition O , where all my cases has one or more of the following diseases: A, B, and C, in addition to my outcome of interest O. So I include individuals with condition A, B and C in my control group as well. In the analysis I add covariates for A, B and C conditions. Now the issue here, its known that risk variants for condition O overlap with conditions A, B and C, which are pleiotropic. So when I adjust for A, B and C, I also adjust for them and as a result I lose those variants in the results. To solve this what I tried is, I did a leave one out analysis.
- GWAS of all samples, adjusting for A, B and C
- GWAS of samples excluding A, adjusting for B and C
- GWAS of samples excluding B, adjusting for A and C
- GWAS of samples excluding C, adjusting for A and B.
I compared the results. Clearly the genome wise hits are different in each case. I can see that the genome wide hits in each of the subset has a p value of at least <0.05 in other subsets. What is the right thing to do here? Anyone had any similar experience and do you have a any statistical suggestion on how to analyse this further?
