Local symmetry in local refinement?


It is possible to apply symmetry in local refinement, but this seems to assume that the symmetry axis of the volume being refined corresponds with the Z axis of the input global refinement. This is often not the case - for example if one has a symmetric homo-oligomer that is a sub-component of a larger asymmetric complex.

Is it possible to apply local symmetry operators during local refinement, and if not would it be possible to add the capabilities (1) to find the local symmetry axis (making the assumption that it intersects the center of the refinement mask) and (2) to apply local symmetry during refinement?


1 Like

Hi @olibclarke,

Thanks for the feature request. In the next release, we will be adding a “Volume Alignment Utilities” job that can translate a volume/mask to a specified origin in 3D, while appropriately modifying the particles’ alignments. The same job will also be able to align the masked volume to the desired symmetry axes.

This can be used for the above workflow by:

  1. Taking the mask to be used by the local refinement, and finding the approximate voxel coordinates of the symmetry center (e.g. if the mask only covers the symmetric homo-oligomer, you could use its center of mass as a guess, from Chimera). This only needs to be approximate because symmetry alignment will search for the optimal residual shift within a specified range.
  2. Taking the above mask, along with the particles/volume from a global refinement, and passing it to the volume alignment utility, specifying the symmetry group, and the above coordinates as the shift.

This will output a new volume, mask, and stack of particles, that have been shifted to the new origin, then aligned to the symmetry axes. These can be used in local refinement with the enforced symmetry. The only thing to note is that symmetry alignment requires that the sub-region actually does have discern-able symmetry (enough to find the location of the symmetry axes), so if the volume is very low-res/doesn’t have clear symmetry, it might be necessary to manually align it to the axes, then re-refine.



this sounds great! is there also a matching utility for making particles cropped around the new center? This would make local refinement dramatically faster for refining small sub-particles of a large assembly…?

1 Like

The easiest way this would be done would be via re-extracting particles from the micrographs, after the above procedure. This would work provided that the micrographs are still accessible, and that signal subtraction hasn’t been done (cropped particles are pulled from the raw micrographs). For that case, we’ll look into a utility that can crop particles without needing to reference the original micrographs.


1 Like

Sounds good! The utility to crop particles would definitely be useful! We are often working with particles that have had bayesian polishing applied in relion

This feature is now running on V3.1, so I was wondering what happens to particles coordinates that are moved outside the volume and is there a way to remove them. I did local refinement on a domain and then centered that portion and did symmetrized local refinement and got similar results before and after centering the domain, but when doing Homogenous Refinement (without symmstry), the structure deteriorated greatly even compared to the original structure taken prior to any localized refinement. I am thinking this is because part of our structure has left the box (we are working on protein lattice) and the particle alignments are no longer valid/useful.

Is there a way to drop these particles? I was hoping Class Probability Filter would be able to recognize particles with alignments outside the box, but even after homogenous reconstruction the alignment scores seem to not me affected by the shift.

The structure extended beyond our box size normally, so we are used to working with mask truncated structures.

Thank you

Hi @schiracha,

Right now there isn’t a tool to remove these particles, but it can be done manually by manipulating the particles’ cs file. You would be looking for particles whose alignments3D/shift (multiplied by alignments3D/psize_A) is greater than half of the box size in Angstroms. More info on manipulating cs files is available on our guide (see tutorial, and additional information on data management). If you see many particles whose alignments are very large, that could explain the discrepancy. Note though that it’s rarely recommended to use homogeneous refinement after local refinement because there is less signal remaining in the structure after masking only part of it, so only local orientation searches can really be trusted.