Methods for Random Modularization of Biological Networks

Biological networks are formalized summaries of our knowledge about interactions among biological system components, like genes, proteins, or metabolites. From their global topology and organization one can learn nontrivial, systemic properties of organisms. In studies of biological network organization empirical networks are typically compared to random network models, and features are identified as important if they are statistically "unusual," i.e. occur surprisingly often or seldom. Naturally, more representative random models result in better feature identification. Since biological networks exhibit a modular structure (mostly pertaining to their hierarchical functional organization), random network models need be modular similarly. In this work we consider the problem of generating random network models that incorporate network modularity. Theoretically, the problem is equivalent to generating random decompositions of a graph into a given number of connected components. Here we describe two methods we have developed to do that and illustrate their utility on pertinent systems biology problems of feature scaling