Abstract:
|
In genome-wide association study (GWAS), variable selection has been used for prioritizing candidate regions that are associated with a phenotypic trait. Especially, in a study of complex traits, we are facing a challenge identifying many regions that are associated with a trait. With the advance of technologies, various large-scale genomic data are more commonly available and combining the information from multiple data seems most promising to enhance our knowledge on genetic associations of complex traits.
We proposed a Bayesian method that can easily incorporate the information of gene networks inferred from gene expressions into the selection of makers, such as single-nucleotide polymorphism (SNP) and methylations. Compared to the penalization methods, Bayesian Lasso, which assumes normal-scale mixtures priors, yields the more robust yet accurate in variable selection. In this study, we proposed the algorithms for Bayesian Lasso in a view of variational inference and extended the model with other priors in a form of normal-scale mixtures. We also worked on a multivariate version of Bayesian Lasso with the application of drug sensitivity tests on cell lines.
|