Functional influence-based approach to identify overlapping modules in biological networks

Abstract

The inherent, dynamic, and structural behaviors of complex biological networks in a topological perspective have been widely studied recently. These studies have attempted to discover hidden functional knowledge on a system level since biological networks provide insights into the underlying mechanisms of biological processes and molecular functions within a cell. Functional modules can be identified from biological networks as a sub-network whose components are highly associated with each other through links. Conventional graph-theoretic algorithms had a limitation in efficiency and accuracy on functional modules detection because of complex connectivity and overlapping modules. Whereas partition-based or hierarchical clustering methods produce pairwise disjoint clusters, density-based clustering methods that search densely connected sub-networks are able to generate overlapping clusters. However, they are not well applicable to identifying functional modules from typically sparse biological networks. Recently proposed functional influence-based approach effectively handles the complex but sparse biological networks, generating large-sized overlapping modules. This approach is based on the functional influence model, which quantifies the influence of a source vertex on each target vertex. The experiment with a real protein interaction network in yeast shows that this approach has better performance than other competing methods. A better understanding of higher-order organizations that are identified by functional influence patterns in biological networks can be explored in many practical biomedical applications.