Multi-Locus Methods for Genome-Wide Association Studies and Genomic Selection

With the development of modern high throughput genotyping technology, an increasingly large amount of genomic polymorphism data have been generated and therefore, genome-wide association studies are becoming a popular tool to identify associations between genomic variants and quantitative traits. Despite the consensus regarding the value of genome-wide association studies for improving our understanding of complex traits, there are well documented problems in the statistical analysis of these data, such as the much larger number of polymorphisms compared to the limited number of subjects, the linkage disequilibrium between polymorphisms, and the confounding issue introduced by population structure, among many others. To meet these challenges, we have developed a series of multi-locus methods, including penalized orthogonal components regression to handle continuous phenotype in association analysis, and generalized orthogonal components regression for case-control studies. Recently, similar methods were developed for mixed models to account for the family structure in family based genome-wide association studies. Furthermore, the method has been successfully applied to genomic selection. On the basis of supervised dimension reduction strategy, the developed methods can handle all polymorphisms across the entire genome simultaneously. We have demonstrated the superior performance of the methods with computer-based simulation studies, and the utility of the methods using real data analysis.