Python-based
Hierarchical
ENvironment
for
Integrated
Xtallography
The purpose of create_alt_conf is to create alternative side-chain conformations for a structure using very high-resolution X-ray data or a very high-resolution cryo-EM map as a guide.
The problem that it solves is that if you have multiple conformations present in a structure, residues that have the same altloc (A or B or C etc) should have plausible relationships, and in particular, the should not clash. Residues with different altloc (A vs B) can have any relationship as they are in different conformers. Finding the assignment of side chain conformations to conformers A, B, C etc that yields good geometry for all the conformers can be difficult.
Note that the model that is produced is not intended to be a final model, rather it is a model that has some plausible alternative conformations. It is recommended that you use the conformers in the output model as suggestions for your model.
The starting point for create_alt_conf is normally a model with a single conformation (no A or B altlocs).
The first step is to generate a set of plausible conformations for each side chain based on the current model and the X-ray data or cryo-EM map.
For X-ray data, a 4mFo-3DFc map is created. This map is expected to show conformations that are not represented in the model. For cryo-EM data, the input map is used.
In either case, plausible conformations for at each side chain position are found by testing each rotamer in a rotamer library for the side chain type present at that position and choosing those that fit the density. The conformer that best fits the density and that is different from the current conformer at that position is chosen. If no suitable alternative is found, the original is used. Conformers that clash with all conformers at any other position are discarded. The new side chain conformers are then used to create a new model with side chains that match the density map and that are as different as possible from the starting side chains. This model is refined using the X-ray data or cryo-EM map. The result of this step is a pair of models, the original and one that has alternative side-chain conformations. Each model has just one conformer.
If you want, you can instead supply a model with multiple conformations and it will just split that model into separate models, one for each conformation (each altloc A or B).
The key step in create_alt_conf is creating a set of models in which each model has good geometry and the models collectively fit the density. The starting point for this step is a set of models that have different side-chain conformations at some or all positions. The challenging part of this step is arranging the side-chain conformers in a way that minimizes clashes but that uses all the supplied side chain conformers.
The procedure used is to generate a diverse set of refined multi-conformer models, to score each model, then to find an optimized model by recombination among the multi-conformer models. The procedure is complicated somewhat by the need to select a small number of test models for scoring, as scoring requires refinement and is therefore a slow process.
The scoring function used to group side-chain conformers is the Holton geometry validation score . This score is a composite of geometry restraints used in refinement and validation metrics and it is a rather good indication of the overall geometric quality of a model. The scoring function requires a refined model, as unrefined models will generally have very poor geometry.
The scoring function is also calculated on a per-residue basis. This per-residue score is the Holton geometry validation score, only including interactions that involve this residue. Note that the sum of per-residue scores calculated in this way does not equal the total score, both because of the way in which scores are calculated, and because some interactions are only within a residue and others are between residues. Even so, the sum of per-residue scores is closely related to the total score. This allows estimation of the total score from per-residue scores.
To reduce the number of models that need to be fully scored, a procedure for probabilistic estimation of the expected score for a model with a new arrangement of side chains is used. In this estimation process, it is assumed that many of the most significant contacts between residues will be with nearby residues (segments of length group_length are considered).
The starting point for this estimation is a set of scored models with varying arrangements of alternative conformations. These arrangments of alternative conformations are set up so that for any stretch of group_length residues, there will be models that match at all other positions but that differ in all possible ways at this set of group_length residues. The differences in per-residue scores among these models are then used to estimate the expected effect of changing the arrangement of conformers for any one residue, in the context of the conformers surrounding that residue. Additionally, an error estimate in this expected effect is estimated.
With these predictors for the effects of changes in alternative conformations at any one residue in the context of its neighbors, an estimate can be made for the expected score for a model with any arrangement of conformations. The score for the most similar model is taken as a starting value, then the minimal set of changes in arrangement is applied and the expected change in score for each change is noted, along with its uncertainty. This yields a probabilistic estimate of the expected score for this arrangement.
The recombination procedure is targeted (not random recombination). The reason for this is that scoring requires refinement, a relatively slow process. The current set of scored models is used as described above to create a probabilistic estimate of the score that would be found for any new arrangement of side chains as conformers in a model. A simple recombination and mutation procedure is used to find arrangements that are predicted to have good Holton geometry validation scores. The arrangements with good predicted scores are then generated, refined and scored to identify optimized arrangements that actually do have good scores.
You can skip the extensive optimization procedure and use a shorter one instead with the keyword quick=True. This is still not that fast, but it is a lot faster than the full version.
If you supply a single model with no alternate conformations, it may contain ligands. Any ligands are separated out and added in (one copy of each ligand for each conformer created) just before refinement.
You can specify anisotropic_refinement=True or supply a single model previously refined with anisotropic refinement. This will instruct create_alt_conf to use anisotropic refinement.
Running create_alt_conf is easy. From the command-line you can type:
phenix.create_alt_conf model.pdb data.mtz
where model.pdb is the model you would like to use as a starting point and data.mtz is an X-ray data file. The tool will run (for a long time) and create a model named model_overall_best.pdb that will contain alternative conformations based on the X-ray data. If you want to use more processors, specify nproc=32 or whatever number you would like. If you want to create more than two conformers, you can specify the number with conformers=3 or whatever you like. If you want to use the conformers present in model.pdb as starting points, rearranging them as needed but not creating any new ones, you can specify use_existing_altlocs=True (or say nothing, this is the default). If you want to throw them away, say use_existing_altlocs=False.
The procedure takes a long time. You would not normally want to run this on a machine with just a few processors. Running with quick=True is much faster, but not as comprehensive.
This procedure creates N full conformers (altlocs A, B, C etc), with all atoms in the macromolecule present in all the conformers. If you want to only have multiple conformations in a few places, you will need to use phenix.pdbtools or another method to remove multiple conformations from the rest of the structure.
This method is only suitable if you have very high-resolution data. Normally 1.5 A is about the lowest resolution data you would want to use.
The procedure only creates alternative side-chain conformers, not main-chain. If you main chain has substantial alternative conformations (not just slight adjustments to match the side chain conformers), you will need to use another approach.