UC San Diego

The dissertation describes computational methods that can be used to predict structure, function, and interactions of proteins by exploiting the wealth of biological databases. The dissertation is divided into four sections. In the first section, a protein interaction prediction algorithmis developed that extends the original protein phylogenetic profiling algorithm. The new algorithm broadens the idea of co-evolution between two whole proteins to include that of their sub-regions. This extension results in identification of co-evolving protein domains that are validated to be interacting. In the second section, the effects of nearest neighboring residues on the backbone-angle propensities of the middle residue are mapped out. An energy function incorporating these effects is tested using threading experiments, and its performance is compared with those of other widely used energy functions. In the third section, the utility of using known structural homologues of a protein fold as a model for backbone flexibility in a protein design algorithm is investigated. Using the design algorithm, 100 protein sequences are designed for each of the structural homologues. Multiple sequence alignments (MSAs) are then derived from the designed sequences for each structure as well as for the entire structural homologues of a protein family. These MSAs are then compared with those of naturally occurring sequences using fold recognition experiments and conservation profiles. Finally, in the fourth section, a method is described that can be used to mine large-scale protein-protein interaction data. By searching protein-protein interaction networks of nine eukaryotic organisms for frequently occurring interaction patterns, previously known protein complexes and pathways are discovered