Structure and evolution of biological networks: A systems biology approach to understanding protein function

In this dissertation I present three significant advances in the theory and practice of systems biology research. The research presented touches upon diverse topics spanning the field, including models for biological network inference, methods for estimating the statistical significance of the results generated by these models, and computational analysis of protein network motif conservation. We conclude by presenting a novel algorithm that integrates heterogeneous genomic data sets to identify key functional regulators of human disease.;First, we implement a robust, efficient and flexible web application (BNET) for the evaluation of computationally and experimentally inferred biological networks. The tool makes use of a powerful statistical network evaluation model and has applications across a broad spectrum of both computational and experimental systems biology research. BNET compares predicted network models against large-scale experimental data mined from public databases, and quantifies the level of evidentiary support for the proposed network in a statistically defensible manner. Applications include rapid evaluation of competing network inference algorithms, consistency and repeatability testing for experimental methods in systems biology, and the assessment of novel coverage achieved by newly developed experimental techniques.;Next, we present a computational evolutionary analysis of protein networks in four species. Our analysis sheds new light on some of the principal assumptions underlying biomedical research in model organisms and leads to the development of an evolutionary model of conservation in protein networks that is consistent with the available network data. Our model additionally supports robust annotation transfer of protein sub-network data from model organisms to human, drastically improving data coverage in the human protein network.;Finally, we develop a novel algorithm for identifying key proteins involved in regulating human disease. This algorithm uses only protein subnetwork information and is therefore able to exploit high-coverage protein networks inferred under our previously described model of sub-network conservation. Case studies on genome-wide diabetes and cancer profile data sets reveal that the algorithm identifies known key regulators of both cancer progression and diabetes.;Taken together, our results represent significant advancements towards the fundamental promise of systems biology: that the development of comprehensive, detailed and robust quantitative network models can drive creation of the biological knowledge necessary to both understand and treat the most intrusive and most costly human diseases.