Analysis of Untyped SNPs: Maximum Likelihood and Single Imputation Methods

• Activity Number: 132
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Yijuan Hu*+ and Danyu Lin
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: association studies ; retrospective likelihood ; single nucleotide polymorphisms ; genotype ; haplotype

Abstract

Although the rapid improvement and reduced cost of high-throughput genotyping have enabled many SNPs to be genotyped in genome-wide association studies, untyped SNPs are still of importance. We examine the properties and performance of two classes of methods. The first class computes maximum-likelihood estimates for models of the joint distribution of the phenotype and the observed genotypes. The second is a two-stage single imputation approach. Simulation studies demonstrated that the maximum-likelihood approach performs well and is generally more powerful than imputation, whereas imputation often yielded biased estimates of effects and underestimation of variances. Theoretical derivations revealed a novel finding that all imputation methods have correct type I error in testing the single-SNP effect. Illustration with a genome-wide association study is provided.