| Popis: |
Population stratification becomes relevant for case-control association studies when allele frequencies are different in cases and controls due to systematic ancestry differences of subjects classified as cases and controls. This may cause spurious associations, and leads to both false positive and false negative findings. Recently, several statistical approaches have been proposed using genomic markers to control for this confounding effect. This study describes a new method that efficiently corrects for stratification by regressing the pairwise genotypic difference on the pairwise genetic distance (computed from genomic markers) of all case-control pairs of subjects. A new test statistic T is formulated to measure the genotypic difference between cases and controls, adjusting for stratification contribution. Significance level is determined by the null distribution of T, which is generated from the genomic markers by using permutation. The current existing approaches (Genomic Control and Structured Association) are compared with this new method by simulating different disease association studies, under a variety of parameter settings. Allele frequencies from the African and European data of the HapMap project as well as simulated allele frequencies from a uniform distribution were used in such simulation experiments. Results suggest that this new procedure has a correct nominal type-1 error rate in the presence of different levels of population stratification. In most scenarios considered, this method has a larger power and, in some cases, substantially larger power than that of existing methods. In terms of power, the Structured Association method is closest to this new approach, but the latter requires substantially smaller computational time, which implies its ease of application in large-scale or even genome-wide association studies. |
