|
Abstract:
|
Complex traits ranging from mRNA expression to disease phenotype are affected by multiple genes, and importantly, by gene interactions (GIs). Gene association networks (GANs) provide a simple, yet effective, summary of relationships among genes. In GANs, nodes are genes and edges represent association often measured by correlation or, as in Gaussian graphical models (GGMs), conditional dependence given all other nodes within a multivariate normal distribution. Although useful, GANs are limited since the association measures quantify a linear relationship among nodes. We propose an approach, applicable for n< p, that uses the correlation and partial correlation matrices to extend the class of GGMs and accommodate GIs. Comparisons between the resulting GI networks and traditional GGMs illustrate the advantages of the proposed approach.
|
