Statistical Topics in Disease Gene Mapping

Abstract

Efforts in disease gene mapping have achieved a great deal of success in mendelain diseases, but made slower progress in common disease studies because of their complexity. The rapid development of genetics and molecular technologies provides an immense amount of DNA data; developing powerful and efficient statistical methodologies is under high demand. This dissertation explored some aspects of the problem. The power of two genome-wide disease gene mapping strategies is investigated. One applies linkage analysis and then linkage disequilibrium (LD) tests to markers within linked regions. The other looks for LD with disease using all markers. The results showed that the genome-wide association based tests are much more likely to identify genes. Genotyping closely spaced Single Nucleotide Polymorphisms (SNPs) frequently yields highly correlated data due to extensive LD, and gives association studies unnecessary and unaffordable burden when these markers don't yield significantly different information. Two procedures are developed to select an optimum subset of SNPs that could be efficiently genotyped on larger numbers of samples while retaining most of the information based on genotypes of a large initial set of SNPs on a small number of samples. One utilizes a spectral decomposition method based on matrices of pair-wise LD, and the other extends David Clayton's htSNP selection method. Properties of the procedures are studied; minimum sample sizes needed for achieving consistent results are recommended; the procedures are evaluated using experimental data. Studying gene-treatment interaction is a long desired problem. When the genetic variation that is being tested is not specific functional sites but randomly selected polymorphisms, a source of randomness is introduced. A mixed effect model is developed to fit fixed treatment effects, random haplotypic effects, and random gene-treatment interactions in this scenario; likelihood ratio tests are applied for testing the random effects. Our simulation results showed that the mixed effect model is valid and generally behaves better than the fixed haplotypic effects model in the exploratory phase of a study.