Ancestry-specific association mapping in admixed populations

During the last decade genome–wide association studies have proven to be a powerful approach to identifying disease-causing variants. However, for admixed populations, most current methods for association testing are based on the assumption that the effect of a genetic variant is the same regardless of its ancestry. This is a reasonable assumption for a causal variant, but may not hold for the genetic variants that are tested in genome–wide association studies, which are usually not causal. The effects of non-causal genetic variants depend on how strongly their presence correlate with the presence of the causal variant, which may vary between ancestral populations because of different linkage disequilibrium patterns and allele frequencies.Motivated by this, we here introduce a new statistical method for association testing in recently admixed populations, where the effect size is allowed to depend on the ancestry of a given allele. Our method does not rely on accurate inference of local ancestry, yet using simulations we show that in some scenarios it gives a dramatic increase in statistical power to detect associations. In addition, the method allows for testing for difference in effect size between ancestral populations, which can be used to help determine if a SNP is causal. We demonstrate the usefulness of the method on data from the Greenlandic population.