Surpisingly low OR - how to diagnose why?

[kyle]

Hello,

After running a GWAS in both plink and plink2, I'm struggling to understand why certain SNPs have very low OR and P, despite the MAF looking similar. 

Below is one example, generated using the code:

plink --bfile case_controls --keep-fam phenotype_list -maf 0.05 --chr 17 --logistic --freq case-control --out unadjusted_test

The output is attached. MAF is 0.45 vs 0.49, in a sample of 2268 cases and 7913 controls. How is it possible the OR is just 0.05?

I tested this in both plink 1.9 and 2, and unadjusted vs adjusted for PCA, the result is consistent.

Thanks,

Kyle

[Christopher Chang]

What is the output of plink 1.9 --freqx or plink 2.0 --geno-counts on this variant?

[kyle]

 Plink 1.9 using --freqx on cases and controls seperately:

Thanks

[Christopher Chang]

When the heterozygosity rate is close to 100%, the "SNP" usually really corresponds to 2 (or possibly more) distinct positions in the original genomes, one of which is usually homozygous for the A1 allele and another of which is usually homozygous for the A2 allele.  Cases like this are usually filtered out with --hwe.

The low OR and p-value are due to the strong association between case/control status and the few non-het genotypes.

[Christopher Chang]

(Forgot to mention, proper analysis of these variants will probably require a more accurate reference genome.)