Detection of Quantitative Trait Associated Genes Using Cluster Analysis

Many efforts have been involved in association study of quantitative phenotypes and expressed genes. The key issue is how to efficiently identify phenotype-associated genes using appropriate methods. The limitations for the existing approaches are discussed. We propose a hierarchical mixture model in which the relationship between gene expressions and phenotypic values is described using orthogonal polynomials. Gene specific coefficient, which reflects the strength of association, is assumed to be sampled from a mixture of two normal distributions. The association status for a gene is determined based on which distribution the gene specific coefficient is sampled from. The statistical inferences are made via the posterior mean drawn from a Markov Chain Monte Carlo sample. The new method outperforms the existing methods in simulated study as well as the analysis of a mice data generated for obesity research.