CONCLUDING REMARKS

Graph-based models of protein interaction networks can combine heterogeneous system-level data sources and their integrative mining can provide valuable information about local and global characteristics of their modular and functional organization. The global structural properties try to characterize the networks as a whole, whereas local network analyses aim at discovering such individual interaction patterns that may carry significant information about their roles in specific cellular mechanisms. Module analyses seek to partition the complex networks into functionally organized hierarchy of interconnected groups that are involved in common cellular functions. While most current network modularization approaches are based solely on the topological properties, such as network hubs, clusters or motifs, it is likely that biologically relevant findings can only be made after contrasting the interaction patterns with other large-scale functional genomics or proteomics measurements. In particular, when mining human interaction networks, that are still limited both in their accuracy and coverage, integrated analysis of interaction networks together with complementary datasets is probably needed to identify robust, functional modules, which may eventually prove valuable also in many medical applications. The limitations of the current protein interaction datasets should be taken into account when developing new data mining methods and when applying these methods to real biological problems. The future success of the module-based analysis is likely to be driven also by parallel improvements both in the experimental techniques for mapping global interaction networks as well as in the modeling frameworks for representing the existing and emerging types of interaction information.