[phenixbb] Max_reasonble_bond_distance error in phene.resl.space.refine

[Rune Thomas Kidmose]

Hi Pavel,

Thx for getting back to me so soon.

I have several examples, however if you are looking for PDB files you can fetch directly from the RCSB etc. then I dont know if I can help.

But some of the files I have piped through my pipeline, always results in the before mentioned error. Both the starting model and map have been fetched directly from the RCSB and the EMDB respectively.

I have included dropbox download links for one such example.

The PDB file: https://www.dropbox.com/s/xi2qlqczmth69bv/last_frame.pdb?dl=0

The corresponding map: https://www.dropbox.com/s/igweq3avgf6o7uk/emd_3765.mrc?dl=0

The example PDB (and map) is a structure of the nucleosome containing both protein and DNA, and the stated resolution of the map is 4.8Å.

I use default settings for the pehnix.real.space.refine run so my run command is:

phenix.real_space_refine last_frame.pdb emd_3765.mrc resolution=4.8

I should mention that it seems we are running phenix version dev-2614 at our department, which I am guessing is not the newest official release?

Anyway I will try and get our sys-admin to install the newest version and then run the same files through phenix.RSR and see what happens.

Thx again for getting back to me

Cheers

/Ruki










On 14 Nov 2017, at 16.20, Pavel Afonine <pafonine at lbl.gov> wrote:

Hi Ruki,

I have been playing around with phenix.real.space.refine as part of a small pipeline I am making for EM structures.

Testing various deposited EM structures and their corresponding maps, within my pipeline, I sometimes run in to the below error:

RuntimeError: Bond distance > max_reasonable_bond_distance: 79.8663 > 50:

I'm trying to chase this down and eliminate. So if you come across a case that can be reproduced please send it to me and I will start from there.

For now simply set "max_reasonable_bond_distance=999" for all your refinements that will mask but not fix the problem. However, this should not affect the refinement results.

Make sure you are using the latest Phenix version.


Followed by a list of distances between, what I am assuming is, atom numbers and the actual distance.

I can get past it simply by changing the max_reasonable_bond_distance for the run to, in this case, 80.

This happens because the procedure optimizes weights by trying a large sample of possible weight values. For some values restraints may be too weak and atoms run one onto another, thus creating a huge clash and strong repulsion that stretches the bond. This leads to the crash.


But I tend to see, that the final results gets significantly worse when I use this bypass.
This is based on very similar models going through the pipeline, where the ones which yields the error, scores much worse when using the above mentioned bypass, compared to the models that showed no max_reasonable_distance_error.

Again, I'm very keep to investigate concrete reproducible cases. So if you have anything to share that would be most helpful!


B: Any ide what I can do, to avoid running into this bond distance problem?

I'd say help us to identify the issue so that we fix it and then you don't run into it any more.

All the best,
Pavel


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20171114/f19d484e/attachment-0001.htm>