Clustering incorporating shortest paths identifies relevant modules in functional interaction networks

Many biological systems can be modeled as networks. Hence, network analysis is of increasing importance to systems biology. We describe an evolutionary algorithm for selecting clusters of nodes within a large network based upon network topology together with a measure of the relevance of nodes to a set of independently identified genes of interest. We apply the algorithm to a previously published integrated functional network of yeast genes, using a set of query genes derived from a whole genome screen of yeast strains with a mutation in a telomere uncapping gene. We find that the algorithm identifies biologically plausible clusters of genes which are related to the cell cycle, and which contain interactions not previously identified as potentially important. We conclude that the algorithm is valuable for the querying of complex networks, and the generation of biological hypotheses.