Inference of ancestral protein-protein interactions using methods from algebraic statistics

Protein-protein interactions are important catalysts for many biological functions. The interaction networks of different organisms may be compared to investigate the process of evolution through which these structures evolve. The parameters used for inference models for such evolutionary processes are usually hard to estimate. This thesis explores approaches developed in algebraic statistics for parametric inference in probabilistic models. Here, we apply the parametric inference approach to Bayesian networks representing the evolution of protein interaction networks. More precisely, we modify the belief propagation algorithm for Bayesian inference to a polytope setting. We apply our program to analyze both simulated and real protein interaction data and compare the results to two well known discrete parsimony inference methods.