Progress in AB Initio Protein Structure Prediction

I will discuss the application of the Rosetta method for ab nitio protein structure prediction to the prediction of protein structure and function on the genome scale. Rosetta is based on a picture of protein folding in which short segments of the chain flicker between different local structures consistent with the sequence and folding occurs when the segments adopt structures and orientations such that low free energy interactions are made throughout the chain. Results from the CASP4 protein structure prediction experiment and larger scale in house tests suggest that for roughly 50% of domains less than 150 residues in length at least one of the top five most commonly generated Rosetta models has sufficient similarity to the native structure to identify it in a structure based search of the PDB. While not suitable for applications such as ligand docking which require atomic level resolution, functional insights can be obtained from Rosetta models through structure based searches of the PDB and characterization of the positions of evolutionarily conserved residues. Rosetta models have been produced for all PFAM families less than 150 residues in length which do not currently have a representative of known structure and are now being generated for proteins in the yeast genome.