Department of Biostatistics and Medical Informatics
University of Wisconsin-Madison USA
Statistical Methods to Map the Genetic Basis of Gene Expression Networks
The reconstruction of biological networks is a key component of efforts to better understand, diagnose, and treat disease. In expression quantitative trait loci (eQTL) mapping experiments, feature selection methods must be employed if networks are to be intepretable, since thousands of phenotypes and genomic loci are available as candidate nodes. Most feature selection methods require that phenotypes map, are correlated with each other or some clinical trait, or share common functional annotation. These requirements have a major impact on the networks reconstructed, and as we demonstrate, can be quite restrictive. I will present an empirical Bayes hierarchical modelling approach for network reconstruction that makes few, and testable, assumptions. The approach allows for quantification of network similarity and difference, and thus provides a phenotype for genomic mapping. Utility is demonstrated in a mouse study of diabetes where eQTL data is available in adipose, liver, islet, and hypothalamus. Our results highlight allele and tissue dependent networks that may provide some functional insight into candidate susceptibility loci identified in recent genome wide association studies for diabetes.