The evolution of protein interactions has produced interaction networks and much biological complexity. Although molecular phylogenetics reveals the end points of evolutionary searches, little is known about the trajectories of interacting proteins through sequence space over evolutionary time. A major bottleneck is the inability to extensively map how binding affinity changes with sequence. A previously developed affinity prediction model will be modified to predict affinity changes upon mutation. Preliminary results show that this method has unprecedented speed and accuracy. In the proposed research, the preliminary models will be improved and thoroughly validated. Subsequently, they shall be used to assess millions of combinations of interface mutations in structurally annotated interaction networks. The mutational robustness of interactions will be determined, compared with theoretical models, and its relationships with observed evolutionary rates investigated at the amino acid, interface, protein and network levels. Then, paths connecting orthologs to each other and to ancestral protein reconstructions will be determined and characterised. The positions on the phylogenetic trees where the affinity imbuing contacts are gained will be determined, as well as the positions where interactions are lost, revealing network rewiring. This will illuminate the changes in interactions over time, and show how interface biophysics constrains the functionally viable paths available for interaction evolution. The relationships between extent of constraint, evolutionary rates and network properties will be investigated. We hypothesise that the inherent ability to form interactions and modulate specificity can explain a number of observations garnered from network evolution studies. Finally, the human-virus exogenous interaction network will be investigated, potentially revealing strategies and counterstrategies employed as viruses vie to hijack regulatory mechanisms.