Side-Chain Prediction Technical Details

The structure refinement program uses the following procedure to re-predict conformations for the side chain set that you select. The default algorithm is as follows:

1. Side-chain rotamers are randomized for nonconserved residues (the default) or for all residues.
2. Beginning with the first residue to be predicted, the side-chain rotamer library is used to find the rotamer with the lowest energy while keeping all other side chains fixed. Once the process is complete for the first residue, the next residue is treated, and so forth until all have been done once (a single pass has been completed).
3. Once the pass is complete, Step 2 is repeated from the beginning, and then repeated again until the side-chain rotamers appear to be converged (no more changes are occurring).
4. Minimization is run on all of the side-chain atoms (but not backbone atoms) of the residues being treated.

Two methods are available for situations where some degree of backbone movement is likely to be required to get the correct side-chain conformation. This typically happens when working with homology models or when performing cross-docking studies, where the backbone conformation of the initial structure might not be entirely correct. These two methods allow for progressively more backbone movement during the side chain prediction calculations, and are described below. They are intended for use on a small set of residues, not the entire structure.

With CA-CB vector sampling

Even minor discrepancies in backbone conformation can result in a slight shift in the location of the CB atom. This slight shift can often cause subsequent side-chain predictions to fail in placing the side chain optimally, particularly on large rigid aromatic residues. Allowing for sampling of the CA-CB vector in conjunction with the side-chain prediction can alleviate this problem. The method is turned on in the input file with the keyword SAMPLE_CBETA. The algorithm is as follows:

1. CB positions in a conical region are systematically sampled around the initial position with a maximum displacement in the CA-CB vector of 30.0 degrees (settable with input file keyword MAXCONEANG).
2. At each position, the optimal side chain rotamer is selected.
3. The best overall combination of CB position and rotamer for the residue is selected from the results for all positions.
4. The sampling iterates over all requested residues until convergence is achieved in the same way as with conventional side chain prediction (see above).
5. Once all side-chain sampling has completed, the side chains and backbone atoms of the predicted residues are minimized.

With backbone sampling

This method is intended for situations where even larger backbone movements may be required, but where a full loop prediction on the entire region may produce more movement than desired. It is turned on in the input file with the keyword SAMPLE_BACKBONE.

For each residue for which side chains are predicted, a backbone loop segment of 3 residues (settable with the input file keyword BACKBONE_LEN), centered about that residue is reconstructed using the normal loop prediction algorithm. Because this loop prediction includes the additional residues on either side of each requested residue, you should expect to see the conformations of these residues optimized as well (i.e. the region that is optimized extends somewhat beyond the residues actually selected for side-chain optimization).

Side-chain prediction is not performed for GLY and ALA residues with any of these methods.