Bayesian Methods for Assessing Gene by Environment Interactions

点击翻译按钮获取中文摘要

Summary/Abstract We propose to develop new statistical methods for studying gene x environment (GxE) interactions using data from molecular epidemiology studies. The focus is on targeted studies, which use single cell gel electrophoresis to measure DNA damage. This technology has great potential for study of GxE, since one can assess how the distribution of DNA damage across cells from an individual varies between experimental conditions. By drawing from cell lines for individuals with known genotype, the NIEHS Comet GxE study seeks to identify single nucleotide polymorphisms (SNPs) related to baseline DNA damage, susceptibility to genotoxic exposures, and repair rate. The phenotype for an individual in such studies is a collection of distributions corresponding to cell-specific DNA damage under different conditions. New methods are needed to efficiently analyze such distributional profiles, while allowing heterogeneity among subjects and SNP selection. The ability to detect GxE interactions is of great public health importance, allowing physicians to better identify patients that are more sensitive to a drug therapy or environmental exposure. Targeted molecular epidemiology studies provide an efficient alternative to traditional epidemiologic designs. Our goals include the following. 1. Develop nonparametric Bayesian statistical methods that allow a distributional profile to vary flexibly across individuals and with predictors, while allowing variable selection. 2. Apply these methods to data from the NIEHS Comet GxE Study to select SNPs associated with baseline DNA damage, susceptibility and repair rates. 3. Develop approaches for including outside information on each SNP, including whether it is in the coding region, is synonymous, is non-synonymous but at a location at which an amino acid change is likely to be damaging, or is in an intron or flanking sequence but is likely to impact gene expression. 4. An additional goal is to develop approximate Bayes methods that can be implemented rapidly, while encouraging sparse modeling of distributional profiles.