Clustering with Overlap for Genetic Interaction Networks via Local Search Optimization

Algorithms for detection of modules in genetics interaction networks, while often identifying new models of functional modular organization between genes, have been limited to the output of disjoint, non-overlapping modules, while natural overlapping modules have been observed in biological networks. We present CLOVER, an algorithm for clustering weighted networks into overlapping clusters. We apply this algorithm to the correlation network obtained from a large-scale genetic interaction network of Saccharomyces cerevisiae derived from Synthetic Genetic Arrays (SGA) that covers ~4,500 non-essential genes. We compare CLOVER to previous clustering methods, and demonstrate that genes assigned by our method to multiple clusters known to link distinct biological processes.