Non-covalent protein-protein interactions underlie most of biological activity on the molecular level. A binding event between two proteins typically consists of two stages: 1) a diffusional, non-specific search of the binding partners for each other, and 2) specific recognition of the compatible contact surfaces followed by complex-formation. Despite significant progress in studying these processes, a number of open questions remain. How do partners find each other in the crowded and interaction-rich cellular environment? What are the exact mechanisms of the specific recognition of binding surfaces? What is the role of induced fit as opposed to conformational selection in the process? We propose to utilize atomistic-level and coarse-grained molecular dynamics simulations and advanced computational techniques in close collaboration with experiment to address these questions, with the ultimate goal of developing a unified picture combining both specific and non-specific contributions to protein-protein interactions. We will focus on several test-cases of broad biological significance, such as the ubiquitin system, to test two central ideas: 1) that protein dynamics is the principal determinant of specific molecular recognition in many systems, and 2) that co-localization, which non-specifically affects the binding process, is a direct consequence of the general physico-chemical properties of the binding partners, irrespective of the features of their binding sites. Methodologically, we will further develop and utilize distributed computing techniques on the world-wide-web and computation on streaming processors to meet the high demand for computational power, inherent in studying protein interactions in silico. In our work, we will closely collaborate with experimentalists, ranging from NMR and X-ray crystallography experts to molecular biologists to both validate our simulations and theoretical work as well as assist in interpreting experimental findings.