Accurate prediction of protein torsion angles using chemical shifts and sequence homology

Torsion angle restraints are frequently used in the determination and refinement of protein structures by NMR. These restraints may be obtained by J coupling, cross-correlation measurements, nuclear Overhauser effects (NOEs) or secondary chemical shifts. Currently most backbone (phi psi) torsion angles are determined using a combination Of J(HNH alpha). couplings and chemical shift measurements while most side-chain (chi(1)) angles and cis/trans peptide bond angles (omega) are determined via NOEs. The dependency on multiple experimental (and computational) methods to obtain different torsion angle restraints is both time-consuming and error prone. The situation could be greatly improved if the determination of all torsion angles (phi, psi, chi and omega) could be made via a single type of measurement (i.e. chemical shifts). Here we describe a program, called SHIFTOR, that is able to accurately predict a large number of protein torsion angles (phi, psi, omega, chi(1)) using only H-1, C-13 and N-15 chemical shift assignments as input. Overall, the program is 100x faster and its predictions are approximately 20% better than existing methods. The program is also capable of predicting chi(1) angles with 81% accuracy and w angles with 100% accuracy. SHIFTOR exploits many of the recent developments and observations regarding chemical shift dependencies as well as using information in the Protein Databank to improve the quality of its shift-derived torsion angle predictions. SHIFTOR is available as a freely accessible web server at http://wishart.biology.ualberta.ca/shiftor. Copyright (C) 2006 John Wiley & Sons, Ltd.