Application of PROSPECT in CASP4: characterizing protein structures with new folds

In the Fourth Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP4), we predicted all 43 targets using our threading application PROSPECT. PROSPECT guarantees to find an optimal alignment between a protein sequence and a structural fold for a general energy function with pairwise contact potential. For each prediction, it gives a reliability assessment based on a neural network approach. In addition, PROSPECT has been added to the Genomic Integrated Supercomputing Toolkit (GIST) and is deployed on terascale computing resources. Structural predictions in CASP4 included three categories, that is comparative modeling, fold recognition, and prediction for structures with new folds. In the fold recognition category, PROSPECT correctly identified 8 of a total of 22 and finished the sixth in the total scores among 127 assessed groups. In the "new fold" category, it found important structural features for most targets, and its overall performance is among the best of all prediction methods. Our CASP4 performance demonstrates that PROSPECT is a powerful tool to quickly characterize structures with new folds, and it may provide useful structural restraints for ab initio prediction methods.