Deterministic Graph-Theoretic Algorithm for Detecting Modules in Biological Interaction Networks

Roger Chang, Feng Luo, Stuart Johnson and Richard H. Scheuermann



Accumulating evidence suggests that biological systems exhibit modular organization. Accurate identification of modularity is vital for understanding this organization. A recent approach, Modules of Networks (MoNet), introduced an intuitive module definition and a clear detection method based on a ranked list of edges generated by the Girvan-Newman (G-N) algorithm. The resulting modules from a yeast network showed significant association with known biological processes, indicative of the method's utility. Despite MoNet's usefulness, systematic bias in the method leads to varied results across trials. MoNet modules also exclude some network regions. Such deficiencies limit meaningful analysis of a network. To address these shortcomings, we have developed a deterministic G-N algorithm (dG-N) and a new agglomerative algorithm, Deterministic Modularization of Networks (dMoNet). dMoNet simultaneously processes structurally equivalent edges while preserving the intuitive foundations of the MoNet algorithm. dMoNet also includes all network nodes in the identified modules, thus generating hypotheses with respect to a full network. A GO-based method for comparison of dMoNet to other modularization methods, including a currently favored algorithm, shows an overall better performance by dMoNet when analyzing large-scale yeast and human protein interaction networks. This comparison method comprises a rational evaluation of the quality of functional modules identified from large-scale interaction networks. The code and resulting data from this work are available upon request.

Index Terms Algorithms, graph theory, interaction networks, modules.

Full Text (PDF)