InterfaceAnalyzerMover fails when interface is defined

[sean-workman]

Hi there,

I am trying to use the InterfaceAnalyzerMover to look at the interface between chains A and C in a PDB file with chains A/B/C.

The script ends up crashing whether I define with the constructor:

ia_native = InterfaceAnalyzerMover("A_C")
ia_native.apply(pose)

or I set the interface after initialization:

ia_native = InterfaceAnalyzerMover()
ia_native.set_interface("A_C")
ia_native.apply(pose)

The script does not crash if I do not explicitly define which interface I want to analyze. I've tried this with my recently solved cryoEM structure, a published cryoEM structure from the PDB that also has these same three chains, and I've also tried with an X-ray crystallographic structure from the PDB that has chains A/B/C/D. I seem to be able to define any interface I like with the X-ray structure without running into issues.

In all cases the structures have been appropriately cleaned (tried both cleanATOM() and clean_pdb.py), but the two structures with three chains (the ones I'm actually interested in) have disulfide bonds. I'm wondering if it is the disulfides that are causing issues.

I have installed PyRosetta with a debug wheel but don't seem to be getting extra information in this case versus the release wheel. When I try running this in a Jupyter notebook I see:

protocols.analysis.InterfaceAnalyzerMover: Using explicit constructor
protocols.analysis.InterfaceAnalyzerMover: Using interface constructor

The Kernel crashed while executing code in the current cell or a previous cell. 
Please review the code in the cell(s) to identify a possible cause of the failure. 
Click [here](https://aka.ms/vscodeJupyterKernelCrash) for more info. 
View Jupyter [log](command:jupyter.viewOutput) for further details.

And when I try running the same code as a Python script rather than in a Jupyter notebook I simply get a segmentation fault.

My PyRosetta version is PyRosetta-4 2024 [Rosetta PyRosetta4.Debug.python312.linux 2024.39+release.59628fbc5bc09f1221e1642f1f8d157ce49b1410 2024-09-23T07:49:48] retrieved from: http://www.pyrosetta.org, but I see the same behavior with the conda installation and when I try with a Docker container.

EDIT1:

It occurred to me that I should also mention that I've tried this with the command line InterfaceAnalyzer using the most up to date conda installation of command line Rosetta and it seems to work fine if I define the interface using:

InterfaceAnalyzer -in:file:s ####.clean.pdb -interface AC

So this does seem to be a PyRosetta specific problem.

EDIT2:

I've now tried looking at different interfaces with PDB IDs 6t2u, 6x6a, and 8tld (all cryoEM) and it seems like the presence or absence of disulfides isn't an issue because all of these fail when the interface is defined. I've also tried with 6co6 and it works just fine (X-ray) when the interface is defined. In all cases the InterfaceAnalyzerMover works just fine if no interface is explicitly defined.

Please let me know if there is additional information I can provide to try and sort out what might be going on here.

[smlewis]

I am intrigued by the thought that this fails in pyrosetta but not C++ rosetta. I find it also strange that you report two invocations of the constructor. I will offer only rank speculation, it has been years and years since I actually looked at any of this (and if it works differently in pyrosetta then who knows what my guesses mean).

You repeatedly mention doing A/C interfaces. The EnergyGraph calculation can meaningfully think about the A/C interface in an ABC pose, but the pull-the-chains-apart method, which is better for most purposes, cannot think about the A/C interface in an ABC pose because what do you do with chain B? I am suspicious that this is a source of your error. You should be able to ask instead about the AB/C and A/BC interfaces and compare them to factor out A/B and B/C to get A/C. Note that if it is shaped more like A--C--B (in other words there's no three part interface) you can just delete B and try that.

You mention disulfides. Are they ACROSS one of your interfaces of interest, or merely present within a chain?

Last thought. Chains in general are an ill defined sort of thing. Sometimes it means "one polypeptide", sometimes it means "one thing that is kinematically grouped through bonds", sometimes people do bad things with PDB files. You said you cleaned the PDBs, I assume AFTER cleaning the chain lettering stays intact as you expected? No surprise chains came into existence?