The Role of Local Ancestry Adjustment in Association Studies Using Admixed Populations

Abstract

Association analysis using admixed populations imposes challenges and opportunities for disease mapping. By developing some explicit results for the variance of an allele of interest conditional on either local or global ancestry and by simulation of recently admixed genomes we evaluate power and false-positive rates under a variety of scenarios concerning linkage disequilibrium (LD) and the presence of unmeasured variants. Pairwise LD patterns were compared between admixed and nonadmixed populations using the HapMap phase 3 data. Based on the above, we showed that as follows: For causal variants with similar effect size in all populations, power is generally higher in a study using admixed population than using nonadmixed population, especially for highly differentiated SNPs. This gain of power is achieved with adjustment of global ancestry, which completely removes any cross-chromosome inflation of type I error rates, and addresses much of the intrachromosome inflation. If reliably estimated, adjusting for local ancestry precisely recovers the localization that could have been achieved in a stratified analysis of source populations. Improved localization is most evident for highly differentiated SNPs; however, the advantage of higher power is lost on exactly the same differentiated SNPs. In the real admixed populations such as African Americans and Latinos, the expansion of LD is not as dramatic as in our simulation. While adjustment for global ancestry is required prior to announcing a novel association seen in an admixed population, local ancestry adjustment may best be regarded as a localization tool not strictly required for discovery purposes.