The correlation along the human genome has allowed GWAS to look for regions associated with disease without having to genotype with all known genetic variants. Although this approach has been successful, it entails that observed GWAS associations will often only be surrogates for the casual variants and will typically represent a noisy measurement of them. One consequence of this is that the disease model as inferred from associated loci may be a distorted version of the true disease model. Through analytical derivations, we have characterized the relationship between disease model parameters and LD, and the resulting impact on power. These show that dominance interaction effects tend to decay quickly, and that such distortions therefore tend to make the disease model look more like a multiplicative model as the correlation between causal and hit SNPs decreases.
