Python-based
Hierarchical
ENvironment
for
Integrated
Xtallography
The routine combine_focused_maps uses rigid-body refinement to place a model in several maps. It then uses the map-model correlation to identify which are the best parts of each map and the relationships among the models in the different maps to superpose the best parts of each map and create a composite map.
Combine-focused-maps uses the fit of chains in a model to each map to identify the correspondence between the maps. Using this approach it is not necessary that the maps are all superimposed or even that they have the same gridding or size.
If the maps all do superimpose (approximately), you can supply a single model. The model will be refined separately against each map with rigid-body refinement. The placements of each chain in each map will then be used to identify the transformations needed to superimpose the parts of each map corresponding to each chain.
If the maps are quite different, then you will need to supply one model for each map, where you have already placed each chain so that it superimposes with the appropriate part of each map. The models should all be the same except that they have been rigid-body-refined against the various maps. You can use the phenix tool phenix.dock_in_map to place the chains in each map if you have not done it in some other way.
Once a set of models has been set up to match the maps, the map-model correlation in the region of each chain in each model is calculated. This correlation is calculated in a special way: the B-values of all the atoms in the model are set to zero before map-model correlation is evaluated. This step is important in order to make sure that a perfect high-resolution map always will score better than a perfect low_resolution map. (If B-values are variable then a perfect high-resolution map and a perfect low-resolution map will both have correlations of 1, but with a B-value of zero the high-resolution map will have a higher correlation.)
For each chain in the model, the map-model correlations for each map are then used to identify the weighting on that map. An empirical weighting scheme is used. The relative weights of two maps (for this one chain) with a difference in map_model_cc of delta_cc are given by:
exp(-delta_cc/delta_cc_norm)
where delta_cc_norm is typically about 0.05. This means a map that has a map-model cc 0.1 less than another map gets a weight of about 1/10 of the better-fitting map. Using this weighting scheme and transformations calculated from the positions of this chain after rigid-body refinement against each map, a weighted average map is created for each chain in the model. These maps are all superimposed on the reference map and masked around the chain that each is to represent.
Finally all the weighted average maps are combined to form a composite map.
Running combine_focused_maps is easy. From the command-line you can type:
phenix.combine_focused_maps reference_map.map focused_map_1.map \ focused_map_2.map model.pdb resolution=4
where reference_map is the map (CCP4, mrc or other related format) that will be used as the template to superimpose all other maps will be superimposed, and focused_map_1.map and focused_map_2.map are maps that are focused on some part of the map (and the remainder of those maps may be poor). The model will be rigid-body refined against all three maps (keeping chains fixed as rigid bodies). The best parts of each map will be selected and combined to create a composite map.
If the density in some of your maps is rather poor, the rigid-body refinement step may not work well. One thing to try in this case is to refine with rigid_body_refinement_single_unit=True. This will tend to hold everything together (but it won't give you an adjustment of individual chain positions). Another option is to refine (with rigid-body refinement only) the models before using combine_focused_maps. That way you have full control over the refinement process. Then in combine_focused_maps use rigid_body_refinement=False.
If you supply a model for each map, the models are used to define what parts of each map are included in the analysis. If the model for your target map has chains ABC and one focused map has just chain A, then only chain A from the focused map will be transferred to the target map. If the model for the focused map has chains ABC then weighted versions of the map in the vicinities of ABC will be included.
If you have a model with symmetry (say, chains A B C D are all the same) and just one focused map (say, focused on A), then you can supply your target map, target model with chains ABCD, an one focused map and a model for the focused map that contains just chain A. Then you add the keyword "use_model_symmetry=True" and the focused map from chain A is applied to A B C and D in the target map.