A0693
Title: Bayesian group Lasso regression for genome-wide association studies
Authors: Lanxin Li - University of Glasgow (United Kingdom) [presenting]
Mayetri Gupta - University of Glasgow (United Kingdom)
Vincent Macaulay - University of Glasgow (United Kingdom)
Abstract: Genome-wide association studies (GWAS) are designed to search across a genome-wide set of genetic variations (SNPs) from different individuals to find SNPs that are associated with a trait of interest. Many statistical methods for GWAS have limitations in accurately identifying SNPs underlying complex diseases (like heart disease), due to weak association signals from SNPs, local correlations between SNPs, and the sheer imbalance between the sizes of the available samples and candidate SNPs. We propose a Bayesian model framework, adapting ideas from Bayesian group Lasso regression, that clusters correlated SNPs into groups, and a population-based MCMC method to conduct powerful group selection in GWAS, to improve the accuracy and efficiency of detecting trait-associated regions. In this model, biological information relating to genomic structure and function can be used to elicit priors that improve the precision of SNP detection; signals from causative SNPs and SNPs correlated with causative ones can be accumulated to make the detection easier; and the total number of variables that need to be tested is vastly reduced. Results from a variety of contexts show that the proposed method improves on a variety of existing methods at association detection, especially when the signals coming from SNPs are weak.
