Using gene expression data to identify causal pathways between genotype and phenotype in a complex disease: Application to Genetic Analysis Workshop 19

Abstract

We explore causal relationships between genotype, gene expression and phenotype in the Genetic Analysis Workshop 19 data. We compare the use of structural equation modeling and a Bayesian unified framework approach to infer the most likely causal models that gave rise to the data. Testing an exhaustive set of causal relationships between each single-nucleotide polymorphism, gene expression probe, and phenotype would be computationally infeasible, thus a filtering step is required. In addition to filtering based on pairwise associations, we consider weighted gene correlation network analysis as a method of clustering genes with similar function into a small number of modules. These modules capture the key functional mechanisms of genes while greatly reducing the number of relationships to test for in causal modeling.