Mixed Modeling of Meta-Analysis P-Values (MixMAP) with Applications to Genome-Wide Association Studies of Low-Density Lipoprotein Cholesterol and Insulin Resistance

• Activity Number: 76
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Gregory Matthews*+ and Andrea S. Foulkes and Muredach Reilly
• Companies: University of Massachusetts and University of Massachusetts and University of Pennsylvania School of Medicine
• Keywords: genetics ; mixed models ; GWAS

Abstract

Genome Wide Association Studies (GWAS) attempt to locate single nucleotide polymorphisms (SNPs) that are associated with phenotypic traits of interest. Results of these studies, which can include millions of SNP locations, are often summarized with a p-value for each SNP indicating the amount of evidence of association. Genes, which often contain many SNPs, are often considered to be detected if any SNP within the gene reaches significance at the genome wide level. This often involves a Bonferroni correction to correct for a large number of tests, resulting in a conservative threshold for detection to maintain a desired family wise error rate (FWER). The MixMAP procedure proposes the use of a mixed effects model and estimating a random effect for each gene which may contain many SNPs. In this manner, genes that contain many strong or moderate, yet not individually significant, SNP signals may be detected. Results from the application of MixMAP to a meta-analysis of low density lipoprotein cholesterol results of over 2 million SNPs and a second study looking at insulin resistance will be presented.