Seminar
Weimin Chen
Efficient Use of Family Data in Genome-Wide Association Scans
With millions of single nucleotide polymorphisms (SNPs) identified and characterized, genome-wide association (GWA) studies are underway to identify susceptibility genes for complex traits and diseases. Both efficient designs and rapid algorithms for data analysis are crucial for a practical genome-wide association scan. In this talk I first illustrate a typical approach for analyzing a genome-wide association scan that includes only unrelated individuals. Then I focus on efficient designs and
analytical strategies for family data. I show that combining high-resolution SNP genotypes for just a few individuals in a pedigree with sparse marker data from a typical linkage scan allows genotypes for
many related individuals to be estimated probabilistically. Given limited genotyping resources, incorporating these estimated genotypes into tests of association can produce substantial increases in power. Either the Elston-Stewart or the Lander-Green algorithm can be used to calculate a probability distribution for each missing genotype, and the resulting probability distributions can be incorporated in a rapid family-based association test. I illustrate the use of this genotype imputation and efficient test on two data sets. The first data set consists of 27 gene expression phenotypes in 20 3-generation families (the CEPH Pedigrees). I show the GWA analysis combining ~800K SNPs available for 90 grandparents and parents with ~6K SNPs available for all 168 individuals. In addition to increasing evidence for association of 15 previously identified cis-acting associated alleles, our genotype inference algorithm allowed us to identify 4 novel cis-acting associated alleles that were missed when
analysis was restricted to individuals with the high-density SNP data only. In the second data set, among ~4,500 extensively phenotyped individuals that were also genotyped at 10K SNPs, I selected ~1,400 individuals to be genotyped at additional 500K SNPs. I illustrate some of the association signals detected using our method.
