IHM crosslink models under the restraint group model are surprisingly not pseudobond groups

[Elaine Meng]

Using "crosslinks histogram" with restraints model number gave "no pseudobonds specified" message when used on IHM model.

For example, I downloaded ​https://pdb-dev.wwpdb.org/static/cif/PDBDEV_00000005.cif
changed name to end in .ihm, opened it in ChimeraX, and then tried command:

crosslinks histogram #1.1.2
No pseudobonds specified.

Model #1.1.2 was a submodel "99 predicted contacts" under Restraints model #1.1. Unchecking the Shown box for #1.1.2 confirmed it is a pbond model. Strangely, unchecking the Shown box for #1.1 did not hide its submodels.

At first I thought the command was failing to recognize pseudobond model number (as opposed to endpoint atomspec) but "crosslinks histogram" worked with both types of spec when applied to hbonds calculated for 1gcn.

I'd initially tried it on a previously downloaded Nup IHM that I'd used before successfully with the command, but when that didn't work, decided to try new IHM from PDB Dev to make sure it wasn't just a stale format problem.

[goddard@…]

The crosslinks models under the IHM “Restraints” heading are not pseudobond groups.  This is very weird.  The idea is that the cross-link appear on multiple models (best scoring, ensemble, starting models) but I wanted a “model” that corresponds to the restraints.  The crosslink restraints model is a special model that knows about those other pseudobond groups and hides and shows them.  This is some crazy experimental stuff along the lines of the “Data Panel” we discussed a year ago.

[Elaine Meng]

Unchecking “S” for #1.1 (Restraints) does not hide anything, so it’s pretty mysterious.

[Tom Goddard]

Turning off display of the Restraint models #1.2.1 or #1.2.2 using Model Panel works correctly with the nup84 example. But turning off the parent Restraints grouping model #1.2 does not work -- the crosslinks are not hidden in that case. That is a bug, currently there is not a mechanism where child models get notified if a parent changed its display state. This could be easily remedied. But the unusual IHM use of models needs to be analyzed and very possibly changed, so I don't plan on fixing this bug until some further analysis of how IHM data should be presented is done. Another example in the IHM dialog of stretching our current concept of what a model is, is that homology models have multiple sequence alignment models (in nup84 these are under Starting Models / NupX / *.ali). These don't correspond to any 3d model and showing them instead brings up a multiple-sequence alignment viewer. These "sequence" and "restraint" models are experimental tests of the Data Panel idea we discussed about 1 year ago.

[Tom Goddard]

Another weird type of model in IHM is that the Starting Model homology templates are not normal atomic structure models. In fact the templates are not loaded and those models are a "load on demand structure". Executing commands on those models as if they are a structure, e.g. display #1.1.1.5 will not display any atoms if the structure has not already been loaded by showing the model with the model panel show button. Ideally any command that might operate on these structures should perhaps load them, but the code is not able to do that. Again this is an abuse of our concept of Model in order to mock-up ideas of how a Data Panel could work.

[Tom Goddard]

Some kinds of data sets stretch the bounds of what is normally considered a model in ChimeraX such as IHM crosslink restraints, X-ray crystallography structures factors (e.g. *.mtz files), sequence files and sequence alignments. The common element is that these cannot be shown alone as a 3D rendering. The crosslinks need a set of end-point coordinates (the data just says residue 12 is within distance 20 Angstroms of residue 344). The x-ray structure factors need phases that often are calculated using an atomic model in order to produce a density map. To open a *.mtz structure factor file using ISOLDE, Tristan requires you to specify an atomic model, and it groups that atomic model with an mtz model. Sequences are not shown in 3D but it is useful to have id numbers like model numbers for them to use in commands.

I experimented with the IHM reader in making crosslink and sequence models that appear in model panel. Basically the idea was to generalize a model to be a dataset. The id number and show/hide buttons (e.g. shows sequence panel) are useful but sequences of course don't have a 3D position in the scene. Obviously it has problems. I think researchers would benefit from having a concept of "dataset" which is more general than our 3d rendered model. But it is complex and I do not have any detailed ideas about how to do this. A few years ago we discussed this idea and talked about a "Data Panel".

I don't expect the ChimeraX IHM experiment to be developing these ideas unless we are able to get funding to support further IHM file format development.