Singling out functional similarities in graph databases

It has been shown that protein-protein interactions analysis may be useful to infer information about biological variations caused by evolution. All the known protein-protein interactions of a given organism may be modelled by a network, namely, the protein-protein interaction (PPI) network of that organism, stored in a graph database. The analysis and comparison of protein-protein interaction graph databases corresponding to different organisms is useful to retrieve information about conservations across species. In this work, we give a contribution in the direction of individuating functional similarities among proteins belonging to different organisms. The technique core consists in computing a maximum weighted matching of bipartite graphs to compare the neighborhoods of pairs of proteins in different PPI graph databases. The idea is that proteins belonging to different organisms should be matched looking not only at their own sequence similarity, but also at the similarity of proteins they “strongly” interact with, either directly or indirectly. Furthermore, the technique allows for the employment of both quantitative and reliability information possibly available about interactions, making the analysis more accurate. We tested the method on the S. cerevisiae and D. melanogaster PPI graph databases, showing its effectiveness in individuating functionally related proteins.