Multiple Imputation in Genetic Studies

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Recently, association studies are being favored over traditional linkage analyses of extended families to determine the genetics of complex diseases. The use of bi-allelic single nucleotide polymorphisms (SNPs) as markers for association/linkage studies has become more common due to their availability and high frequency throughout the genome. With the advent of such new technologies as the Perlgen chip, it is possible to genotype about 1.6 million SNPs on the whole genome on each individual. However, this creates a new set of problems and challenges in analyzing data with missing values. Missing values are inevitable in genotypic as well as phenotypic data, and clearly incomplete data could give rise to undependable and/or biased results in gene mapping. The missing data problems in familial data with phenotypes and the large number of available genotypes have not been adequately addressed. Most statistical analysis software uses list-wise or pair-wise deletion that can potentially create a problem by deleting most of the data for the analysis thus resulting in serious bias. Currently, there is no software available for Multiple Imputation for familial or case-control genetic data. The overall goal of this proposal is to provide a bridge between, on the one hand, these well-established techniques for estimation in the presence of missing data and, on the other hand, the most commonly used statistical analysis methods of genetics data. In particular, the Specific Aims of this proposal are: Specific Aim 1: To model and simulate multivariate phenotypic and genotypic data for family and case-control design based on real data. Specific Aim 2: To compare missing data procedures for robustness and reliability using simulation and then to develop Multiple Imputation software.