B0252
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)
Indranil Mukhopadhyay - Indian Statistical Institute (India)
Abstract: Genome-wide association studies (GWAS) have become the most commonly used experimental design to detect association between a trait of interest and genetic variation across the genome in the form of single nucleotide polymorphisms (SNPs). However, many statistical methods for GWAS have limitations in accurately identifying SNPs' underlying traits related to complex diseases, due to the weakness of association signals, local correlations between SNPs in particular genomic regions, and the sheer imbalance between the size of the available sample and the number of candidate SNPs. A Bayesian framework is proposed, adapting ideas from group Lasso regression, that seeks to detect groups of correlated SNPs associated with the trait more accurately. In this model, priors are informed by biological assumptions about the sparsity of associated groups to improve the precision of association detection; signals from causative SNPs and SNPs correlated with causative ones are accumulated to make the detection easier; and the total number of variables that need to be tested is vastly reduced. A population-based MCMC method is used for efficient posterior sampling. Results from a variety of contexts show that the proposed method improves on a variety of existing methods at association detection, especially when signals are weak.