|
Abstract:
|
Several statistical issues arise in genome-wide studies of association between a phenotype and individual single nucleotide polymorphisms (SNPs). These include large scale multiplicity of tests, possibilities of spurious or attenuated association due to population structure (i.e., hidden stratification), need for different statistical models depending on the phenotype's mode of inheritance at each SNP, and adaptability of techniques to various sampling designs such as cross-sectional or case-control. We present a framework using Bayesian Nonparametric models, with emphasis on Dirichlet process mixtures, that addresses these issues. Implementation of the framework is illustrated in specific cases and compared with other widely used methods available in each.
|
