Loop Refinement Technical Details

Loop refinement using the Default option proceeds as follows:

The loop is reconstructed using the backbone dihedral library, by building up half from each direction. The resolution starts off very coarsely, becoming finer until it manages to produce a required number of physically realistic loop conformations (based on the length of the loop).

The large number of loops generated in the first step are clustered, and representatives of each cluster are selected.

These cluster representatives are then scored as follows:

Side chains are re-added to the loop residues, as well as any residues within a specified cutoff, according to the side-chain optimization procedure.

The loop and contact side chains are minimized.

The energy is calculated.

4. The best scoring loop structures are returned.

Loop prediction with extended sampling is specifically designed to overcome sampling problems with long loops, in which the number of residues results in an unwieldy number of possible loop conformations.

Extended sampling uses the following procedure:

Two initial predictions are run on the loop. These are intended to coarsely sample conformational space and are carried out with two slightly different sets of prediction parameters. Each initial prediction returns up to four top-scoring conformations, resulting in up to eight structures. These eight structures are intended as initial points for finer sampling of what appears to be favorable regions (basins) of conformational space.

Each of the eight structures generated in Step 1 is run through another stage of refinement, with a 4 Å restriction on Cα movement. This is intended to sample the basins more finely.

The eight refined structures from Step 2 are combined with the original structures from Step 1 , and this set of 16 structures is ranked by energy. The five top-scoring structures are then subjected to a final stage of refinement, with a 2 Å restriction on Cα movement. This is intended to perform fine-grained sampling of the best basins.

This procedure defines the “Extended High” sampling procedure. The “Extended Medium” and “Extended Low” carry four and two structures, respectively, into Step 2 and Step 3 .

In ultra-extended sampling, the three steps of extended sampling are followed by the steps below.

The top four structures are selected, and two predictions run on each, with part of the loop frozen. Runs are done with “moving windows” with varying numbers of contiguous residues: all n residues, the two possible contiguous sets of n-1 residues, the three sets of n-2 residues, and so on. The structures are sorted after each cycle of a given number of residues.

Step 2 is repeated with the top eight structures cumulatively generated, then Step 3 is repeated with the top eight structures cumulatively generated.