1. Sjors Scheres Avatar Sjors Scheres
  2. Untitled project
  3. relion-3.0_beta
  4. Issues

Weirdness using solvent-flattened FSCs in masked refinement

Issue #7 closed
Oliver Clarke created an issue

Hi Sjors,

I am seeing something odd when performing a masked refinement with solvent-flattened FSCs.

Basically, auto-refine never progresses to higher resolution, despite the fact that the FSC is near 1 at the resolution limit. When I run the refinement with the same mask but with solvent-flattened FSCs off, I don't observe the same behavior (screenshots of the relevant parts of _model.star files attached).

I have run refinements with solvent-flattened FSCs on other datasets without issue using relion 3beta, so I think it is something specific to this dataset, but this still seems like a potential bug unless I'm missing something. Let me know if there are any files I can provide that would be helpful to debug if so.

Cheers Oli

Comments (7)

  1. Oliver Clarke reporter

    This is definitely happening for other datasets too - it only seems to happen if I start (or restart) refinement from local searches (e.g. healpix order 4), if that's any help. E.g. I have one refinement where I start refinement from 1.8deg searches, and refinement proceeds to an estimated accuracy of angles of 0.7deg & Pmax of 0.9, but relion refuses to increase the angular increment or the resolution beyond which phases are randomized. Happy to provide data or logs as needed

  2. Sjors Scheres repo owner

    Hi Oli, Could you run postprocessing on that iteration unfil.mrc maps? That will yield insights in what's happening with the phase randomisation, which may be useful in finding out what happens. Thx, S

  3. Oliver Clarke reporter

    Hi Sjors - sorry for the delay! See attached postprocessing FSCs and output for identical runs with and without solvent_correct_fsc (both with the same mask)

    Cheers Oli solvent_correct_post.pngsolvent_correct_fsc.pngno_solvent_correct_post.pngno_solvent_correct_fsc.png

  4. Sjors Scheres repo owner

    Hi Oli, I think your mask is too tight. Could you try and run the refinement with a softer mask, you basically want to bring that red line down, just like you would in normal post-processing. HTH, S

  5. Oliver Clarke reporter

    Hi Sjors - I don't think the mask is too tight - it has a soft edge of 9 pixels, and I have seen the same behavior with other datasets. I can try to find another example?

  6. Oliver Clarke reporter

    I agree this is probably not the best example because the masked area is very small - I will try to find an example where I have a direct comparison with relion 2.1 - my impression is that the behavior of solvent_corrected_fsc has changed

  7. Oliver Clarke reporter

    Haven't had time to compare with RELION 2.1 yet... will close and re-open a new issue if/when I have a direct comparison that clearly shows a problem. Sorry for the bother!

  8. Log in to comment
Assignee
–
Type
bug
Priority
major
Status
closed
Votes
0
Watchers
1
Jira: the preferred issue tracker for Bitbucket. Join the team!