Sub-particle extraction and localized recon

Is it possible to extract sub-particles from an Icosahedral reconstruction of a virus? I am referring to virtually any area on the viral surface and not just around symmetry axes or just the asymmetric unit. Do have to provide a mask for the region of interest in the volume alignment step, with the new center being the center of the mask?

Thank you.


Yes, you can do this. Run symmetry expansion, then volume alignment tools with your desired mask and new center. Then when you extract using the coordinates out put by volume alignment tools, you will extract each subparticle in the virion.


1 Like

Thanks Oli. I will try it and certainly share my experience.



It’s easy to do, but I usually do this sort of thing in RELION because for the datasets I’ve experimented with, CryoSPARC converges too quickly and tends to leave some resolution “on the table”. RELION can (and usually will) continue to refine down to 0.02 degrees angular sampling, resulting in reconstructions which can be as much as 3 Å higher resolution for really, really big objects; even with Ewald sphere correction enabled.

However, whether it will give you an appreciable improvement in resolution or map quality will depend greatly on how large your virus is.

1 Like

Well, I typically use Scipion for this purpose. But I was wondering whether this could be done in cryoSPARC. So far I have always used Relion for the 3D classification followed by refinement. I have never gotten to resolution ranges beyond 3.5A with the kind of samples I work with. Even the best parts of the glycoprotein shell is around 3.5 A. I will try cryoSPARC and check the output.


Have you tried local refinement in cryoSPARC after global refinement (e.g. after NU refine)? It generally gives a bit of a bump, and allows specification of the maximum angular sampling. This can be useful in cases where non-uniform refinement is beneficial (although SIDESPLITTER, a plugin for relion, can give similar results in some cases).

Yup, tried local refinement. It does help a little. The larger the object the less it helps, although there is a definite balance between box size for the blocks/focussed segments. Too small and it gets all confused and angles drift off (same happens in RELION), too large and there is little to no appreciable increase in processing speed (although obviously box sizes drop so crazy mega-systems aren’t so essential).

I need to test SIDESPLITTER more…

1 Like

Tested the sub-particle extraction and recon. Works nice. Have to try with some high res data to see how good the classification is. I have tried sub-particle extraction after NU refine (icosahedral recon) followed by 3D classification and refinement. In this particular case, the resolution was slightly better. Although, technically, it was not local refinement. Haven’t tried SIDESPLITTER, need to try it.



@rbs_sci Does CS local refinement still do poorly if you set these parameters?

  • Pose/shift priors ON
  • Pose/shift std reasonable values (like 7˚, 4Å for a high resolution refinement, maybe just think about your object size/shape in relation to this)
  • Leave search ranges as 3x std
  • Recenter poses and shifts ON
  • Set angular sampling to a smaller value, e.g. 0.1 or 0.05
  • Initial low pass is reasonable, like 4.5-6Å for ~3Å refinement, maybe lower like 12Å if the object is large and well-aligned
  • Test with and without marginalization and nonuniform refinement
1 Like

Will test and get back to you on a run with those parameters.

Is there a tutorial/protocol/video for this? Similar scenario, I have an icosahedral virus with regions of localized asymmetry. Thanks in advance.

I’m not sure re a tutorial - but have you tried the general procedure in this post? Did you run into trouble with one of the steps? Can explain further if needed