Health Research and Policy


DATE: December 6, 2012
TIME: 1:15 - 3:00 pm
LOCATION: Li Ka Shing Center for Learning (LKSC)
291 Campus Dr, Room LK 209
1:15 pm - 3:00 pm
TITLE: Model Selection Approach for Genome Wide Association Studies in Admixed Populations
SPEAKER: Malgorzata Bogdan
Associate Professor at the Departments of Mathematics and Computer Science, Wroclaw University of Technology and Jan Dlugosz University in Czestochowa, Poland
Fulbright Visiting Scholar at the Department of Statistics, Stanford University
Genome Wide Association Studies (GWAS) are used to identify regions of the genome hosting genes influencing traits of interest. During such studies scientists test a large number of genetic markers for the association between their genotypes and a given trait. This creates a huge multiple testing problem and results in a relatively low power of detection of influential genes. In admixed populations, which originate from a recent interbreeding between two previously isolated populations, one can locate influential genes by using admixture mapping, where the information on the genotypes of genetic markers is replaced with the information on the ancestry of a given region of the genome (see e.g., [3] and [4]). Due to the strong correlation between the ancestry states at neighboring genome locations the multiple testing correction can be substantially relaxed in comparison to the classical GWAS. This is however counterbalanced by a non-perfect correlation between the genotypes and the ancestry states. In recent articles some methods for genome wide association studies which combine the information on the genotypes and ancestry were proposed (see e.g., [4]). These methods rely mainly on single marker tests. However, it is known that the power of detection of causal genes in GWAS can be substantially enhanced by using linear models, which allow to estimate the joint influence of several genes (see e.g., [2]). In this talk we will show how the model selection approach to GWAS based on the modified versions of the Bayesian Information Criterion (see e.g., [1] and [2]) can be extended to use the information on the ancestry states. Our simulation studies show that the ancestry information can help to detect influential genes in the regions of low linkage disequilibrium and, due to elimination of these genes from the residual error, can increase the overall power of detection of other genes.

This is a joint work with Hua Tang from Stanford University, Florian Frommlet from Medical University of Vienna and Piotr Szulc from Wroclaw University of Technology.

Suggested reading:

Back to Workshop in Biostatistics -->

Stanford Medicine Resources:

Footer Links: