Spectral Graph Theory and Ancestry in Genome-Wide Association Studies

• Activity Number: 296
• Type: Invited
• Date/Time: Tuesday, August 4, 2009 : 10:30 AM to 12:20 PM
• Sponsor: IMS
• Author(s): Kathryn Roeder*+ and Ann Lee and Diana Luca
• Companies: Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University
• Keywords: spectral analysis ; high dimensional ; association ; principal components ; genetics ; genomics

Abstract

Data for genome-wide association studies are being collected for a myriad of phenotypes. Dissecting ancestry based on principal component analysis (PCA) is the traditional approach to summarizing patterns of ancestry, but a number of problems can arise. PCA is quite sensitive to outliers and fails to detect the key axes of variation when spurious data are present. Moreover, if the sample is ascertained from individuals of disparate ancestries, PCA detects the major axes of variation, but often fails to detect subtle ancestry. As an alternative to PCA, spectral analysis can overcome many of these obstacles. We develop a statistical method to find the axes of genetic ancestry that is not sensitive to outliers and that also discovers major and subtle substructure. We apply these methods to several GWA studies to illustrate the approach.