Meta-analysis of high-throughput gene expression data is often used for the interpretation of proprietary gene expression data sets. We have recently shown that co-occurrence patterns of gene expression in published cancer-related gene expression signatures are reminiscent of several cancer signaling pathways. Indeed, significant co-occurrence of up to ten genes in published gene expression signatures can be exploited to build a co-occurrence network from the sets of co-occurring genes ("co-occurrence modules"). Such co-occurrence network is represented by an undirected graph, where single genes are assigned to vertices and edges indicate that two genes are significantly co-occurring. Thus, graph-cut methods can be used to identify groups of highly interconnected vertices ("network communities") that correspond to sets of genes that are significantly co-regulated in human cancer. Here, we investigate the topological properties of co-occurrence networks derived from published gene expression signatures and show that co-occurrence networks are characterized by scale-free topology and hierarchical modularity. Furthermore, we report that genes with a "promiscuous" or a "faithful" co-occurrence pattern can be distinguished. This behavior is reminiscent of date and party hubs that have been identified in protein-protein interaction networks.
Keywords: PubLiME; co-occurrence network; date hub; hierarchical modularity; party hub; scale-free network.