Prediction of protein-protein interactions and function in bacteria

In recent years, several bacteria-specific methods for predicting protein-protein interactions have been developed. The most widely used of these are: gene cluster (GC), gene neighbor (GN), rosetta stone (RS), and phylogenetic profile (PP). These methods have been shown to be powerful tools. I provided guidelines for when each method is appropriate by exploring different features of each method and devised potential improvements offered by their combination. Overall, combining interactions from all sources allowed better function prediction when the sources and interactions were weighted by their reliability versus using a simple unweighted union of the interaction sets.;Next, I explored different features of phylogenetic profile (also called co-conservation). Phylogenetic profile is a well-established method for predicting functional relationships between proteins. I examined how various aspects of this method affect the accuracy and topology of protein interaction networks. I showed that the choice of the reference genome in this method influences the number of predictions generated involving proteins of previously unknown function, the accuracy of predicted interactions, and the topology of predicted interaction networks. A major limitation of cluster co-conservation (CCC) is that it has previously been limited to finding interactions within a single target species. Here, I have extended CCC to develop protein interaction networks based on co-conservation between protein pairs across multiple species. This method is called cross-species cluster co-conservation (CS-CCC). The results showed that CS-CCC provides unique and biologically useful information that is not identified using CCC alone. Moreover, CS-CCC can be used to help uncover systematic errors in annotation and facilitate the blind annotation of new genomes. I hypothesized that in bacteria, the topology of protein interaction networks derived via co-conservation information could improve methods for predicting protein function. The results showed the co-conserved protein-protein interaction networks had scale-free topologies.;The results indicated that some properties of the physical yeast interaction network also hold in the bacterial co-conservation networks, such as high connectivity for essential proteins. However, the high connectivity among protein complexes in the yeast physical network is not seen in our bacterial networks. The distributions of node connectivity were varied by protein functional category and could be informative for function prediction. By integration of functional information from different annotation sources and using the network topology, I was able to infer function for uncharacterized proteins.