The goal of this project is to develop novel computational techniques to significantly improve the state-of-the-art in four areas of protein structure prediction: (1) protein fold recognition capable of identifying structural homologs when the sequence similarity is within or below the so-called """"""""twilight-zone"""""""";(2) empirical potential functions that can effectively identify the best models from those that are problematic and can guide the generation of high-quality structural models;(3) accurate sequence-structure alignments, through expanded structural rules, through simultaneous backbone threading to take full advantage of the more accurate two-body and three-body energy functions, and through systematic and rapid generation and application of limited structural data from experiments;and (4) accurate prediction of loops and side-chains through applications of novel loop generation and prediction techniques, and through development of rigorous and efficient algorithms for side-chain packing prediction. The key elements of the outcome of this project will be an ensemble of very effective prediction and modeling computational methods and implementations, which can address the most challenging problems in building high quality structures from sequences. A computational pipeline for high accuracy structure prediction designed for proteins that do not have close structural homologs in PDB will be developed. These computational capabilities could have profound impacts to drug design and disease studies.
Showing the most recent 10 out of 31 publications
