Searching for disease susceptibility variants in structured populations

Abstract

Data for genome-wide association studies are being collected for a myriad of phenotypes. Many of these studies do not include control samples selected to reflect ancestry similar to the case samples. At the same time "control databases" are becoming available to be utilized as a common resource. These data are often genotyped using a large-scale SNP array. Human populations exhibit complex structure that can lead to spurious associations if not properly handled. How to couple case and control databases effectively is a pressing question. We review available methods for modeling genetic ancestry based on the information gleaned from the SNP array. Methods for selecting control samples with genetic ancestry similar to the case samples are described.

Figure 1

(a) Simulated data with nine subpopulations following a continental cline. One axis is required to capture the gradient structure. (b) Three subpopulations with admixture. The first two eigenvectors are needed to show the structure.

Figure 2

(a) EVD applied to AIM markers reveals one outlier. (b) First two eigenvectors for AIM markers after removing the outlier; short (red) and tall (black) appear uniformly spread across the clusters. (c) GEM detects 6 unmatchable observations (blue).