Macromolecular Modeling Blog ™

Roland Dunbrack shows off the capabilities of SCWRL4, the new version of his famous side chain prediction software. Indeed it seems that the new version has come a long way, especially in the form of a new, smoother backbone-dependent rotamer library. Roland promises to answer any questions regarding the new rotamer library, SCWRL, and also random questions.

We are releasing SCWRL4.0, a new version of our side-chain conformation prediction software.

SCWRL4 is more accurate than SCWRL3 and has a number of improvements:

• a new backbone-dependent rotamer library with more detailed treatment of non-rotameric degrees of freedom (Asx, Glx, aromatics).
• a hydrogen bonding function
• a soft van der Waals energy term
• a tree decomposition algorithm that enables solving problems with a longer-range energy function and more conformations
• a heuristic algorithm for simplifying and solving dense graphs
• sampling about rotameric conformations (“sub-rotamers”)
• optionally solving the prediction problem within the crystal
• rigorously evaluated

SCWRL4 is available for both Windows and Linux.

Users at non-profit institutions may register for a license here.

Users at commercial institutions should contact Roland.Dunbrack@fccc.edu for commercial licensing information.

The paper is in press: Krivov, Shapovalov, and Dunbrack. Proteins: Structure, Function, Bioinformatics, 2009. A preprint is available at the website above.

Two features of the paper that may be of general interest include:

1) calculation of side-chain conformations in the crystal by extending the graph to include edges between residues that are far apart in the structure but interact with each other in the crystal. That is, SCWRL4 can make bona fide predictions of side chains within crystals.

2) calculation of prediction accuracy as a function of electron density and side-chain burial in the crystal. For side chains that are fully buried in the crystal (accessible surface area = 0), which includes side chains in crystal, oligomer interfaces and in monomer cores, SCWRL4 predicts 95.4% of chi1 and 91.1% of chi1+2 correctly within 40 degrees. At the highest electron density, SCWRL4 predicts 95.5% of chi1 and 88.8% of chi1+2 correctly. We show accuracy vs electron density curves as well as accuracy vs surface accessibility curves. 

By Roland Dunbrack.

Roland Dunbrack’s group develops methods and software for protein structure prediction, including the backbone-dependent rotamer library, SCWRL, and MolIDE.