| Networks have been an invaluable framework for modeling biological systems. It is the proteins that carry out almost all biological processes, and they do so by aggregating with other proteins instead of acting alone. Hence, analyzing protein-protein interaction (PPI) networks could provide insights into underlying cellular processes. Analogous to genetic sequence research, biological network research is expected to lead to an explosion of knowledge about evolution, biology, and disease. We present novel algorithmic solutions for topology-based alignments, analyses, and modeling of biological networks, hence strengthening the link between network topology and biology. The importance of our methods is that they extract biological knowledge from a new source of biological information, pure network topology, independently of any other data source.;We introduce two novel algorithms for global network alignment that result in the most complete topological alignments to date. We use our alignments to transfer biological knowledge across species and reconstruct their phylogeny, suggesting that network topology represents a valuable source of biological and phylogenetic information.;We further link biological function with PPI network topology. First, we demonstrate that topology can be used as a complementary method to sequence-based approaches for homology detection. Second, we show that topology around cancer genes is different than topology around non-cancer genes and use this observation to predict novel cancer genes. Third, we couple our topology-based computational methods with functional genomics to identify novel members of known melanogenesis-related pathways, validating our predictions phenotypically. Fourth, we find that "topologically central" human proteins correspond to functionally important genes, e.g., aging, cancer, and pathogen-interacting genes. Our methods could lead to identification of novel drug targets, thus aiding therapeutics.;Finally, we demonstrate that the choice of a network property for evaluating the fit of a model to the data is non-trivial, since different models might be identified as optimal with respect to different properties. Also, we design a new, mathematically rigorous network property that imposes a large number of similarity constraints on networks being compared and use it to find optimal model for PPI networks.;In summary, we provide compelling evidence that important biological knowledge can be learned from network topology. |
