The long-term goals of this proposal are to predict protein structures from sequences and to predict mutation-induced changes in protein stability using bioinformatic, theoretical, and computational methods. This proposal addresses the challenge of improving existing knowledge-based energy functions that describe the interactions among amino acid residues and between amino acid residues and the aqueous environment. We propose to develop a new approach that employs the principles of statistical mechanics, rather than statistical methods to derive knowledge-based potentials of mean force from know structures. Preliminary studies show that the new approach leads to an all-atom distance-dependent pair potential that is significantly more accurate in structure selections from decoys and stability prediction than two previously developed all-atom knowledge-based potentials. This initial success provides strong incentives for the further development and validation of this approach. More specifically, the new approach will be extended to extract the backbone torsion and three-body potentials and to take into account the solvent effect more explicitly. The accuracy of the potentials developed will be tested by structure selections from multiple decoy sets and the prediction of mutation-induced changes in stability. The successful completion of the proposed studies will likely lead to a new generation of algorithms for more accurate structure determination.
Showing the most recent 10 out of 39 publications
