Statistical Methods for Constructing Weighted Gene Coexpression Networks, Applications to Identifying Complex Disease Genes in Mouse Crosses

• Activity Number: 470
• Type: Invited
• Date/Time: Thursday, August 11, 2005 : 8:30 AM to 10:20 AM
• Sponsor: WNAR
• Author(s): Steve Horvath*+ and Bin Zhang
• Companies: University of California, Los Angeles and University of California, Los Angeles
• Keywords: scale free topology ; social network analysis ; systems biology ; linkage mapping ; weighted networks ; complex disease

Abstract

Microarray gene expression data allow one to define gene coexpression networks, which approximately satisfy a scale-free topology. Recently, several authors have found scale-free topology has important implications that metabolic networks (i.e.. highly connected "hub" genes) are essential for yeast and Celegans. We describe statistical methods for constructing weighted gene coexpression networks and propose a statistical mixture model for such networks. We generalize the notion of topological overlap between two genes and use the resulting dissimilarity measure to detect gene modules (sets of tightly correlated genes). We propose to use multidimensional scaling methods to visualize such networks. We use principal component analysis to characterize highly connected "hub" genes and for understanding the relationship between gene modules. We find the connectivity of a gene in the network can be used to predict its relevance for clinical traits of interests. We outline a strategy for using these coexpression networks on mouse crossdata to identify complex disease genes. This is joint work with Bin Zhang, Jake Lusis, Tom Drake, and Eric Schadt.