Detecting Network Motifs in Gene Co-expression Networks Through Integration of Protein Domain Information

Abstract

Biological networks can be broken down into modules, groups of interacting molecules. To uncover these functional modules and study their evolution, our research groups are developing graph-theory based strategies for the analysis of gene expression data. We are looking for groups of completely connected subgraphs (e.g., cliques) in co-expression networks in which corresponding members (genes) encode proteins with the same combination of protein domains. The common pattern shown by a group of such cliques is a “network motif” that may be reused multiple times within organisms. We have developed algorithms for constructing gene co-expression networks labeled with corresponding protein sequence domain combinations, and then detected recurring network motifs with similar protein domain memberships within these labeled networks. The statistical significance of detected network motifs is evaluated by comparing results with those from randomized networks. Also the biological relevance of network motifs is evaluated for shared Gene Ontology annotations on biological processes. We applied our approach to the malaria transcriptome and found many network motifs with three, four, or five members. Many predicted network motifs were further supported by their existence in yeast protein interaction networks. These results illustrate a new strategy for studying the modularity of biological networks by integrating different types of data and cross-species comparisons. A full description of results is available at http://mouse.ornl.gov/~xpv/camda04/.