The protein is a homo-tetramer of ~300 kD. I completed 2D classification-ab initial reconstruction-heterogenous refinement. Only one good 3D class was identified. I did non-uniform refinement of this one 3D class and got 3.1 angstrom resolution with C1 symmetry enforced.
This is a small molecule-protein complex. I want to know if all four subunits are bound with the small molecule. I did 3DVA and 3DVA Display (cluster option). I removed duplicate particles from particle clusters from 3DVA. Then I tried to refine volume cluster from 3DVA using non-uniform refinement (C1). The following warning appeared.
“Symmetry expanded particles were detected. Are you sure you want to use symmetry expanded particles in a refinement with global orientation search? This may result in duplicated superimposed particles and may invalidate the gold standard split.”
Please suggest what refinement I should use on the cluster volumes from 3DVA.
You should not use global search after symmetry expansion. Try local refinement with a static mask around one monomer.
You could also try particle subtraction (remove protein signal) and 3D classification without realignment, with a mask including the three active sites. This trick often works very well in relion, I have not tried that approach in cryosparc.
Agreed in general - although if @miles has performed remove duplicates, that effectively undoes symm expansion, so he could do NU-refine or other global searches, and sometimes this is useful.
However if you have a pseudosymmetric particle global searches will probably undo the good work you did by 3DVA/classification - so yes you usually want to proceed with local refinement, without removing duplicates.
I think the replies above are absolutely on-point! I’ll provide a bit of background on this warning for completeness’s sake.
As @peter.cherepanov said, in general it is best to avoid refinements with global pose searches (Non-Uniform and Homogeneous Refinement) with expanded particles. This is because, if half-sets are reassigned, duplicate particles may end up in the opposite set and with the same pose, which would break the GSFSC assumptions and give a spuriously high resolution estimate.
In your case, if duplicate particles have been removed, you are left with only a single copy of each image, so this warning can be ignored, as @olibclarke said. However, the information from symmetry expansion is still there (specifically, the sym_expand fields in the dataset). Therefore, we still present this warning since we do not know that there is again only one copy of each image.
As an aside, this exact case is why we decided to make this a warning rather than fail the job outright.
It’s an interesting point about half sets. Would it not be safer to leave half sets as they were (so that triplicated particles retain their original half set identity)? I do not know how cryosparc does it, but symmetry expansion can be done with other tools like star.py. After that, the latest cryosparc versions will not reshuffle half sets by default during NU refinement.
Still, global refinement will just realign the expanded particles back to no benefit.
Not always to no benefit - for example, consider you have an oligomeric protein with n independent binding sites.
You want to identify particles where all n binding sites are occupied. You can run classification with a mask around the binding site after symmetry expansion, select just those particles which have all n symmetry copies in the same class, and remove duplicates so you can refine a “full-holo” set with point group symmetry. There are other such cases where this workflow can be useful - but totally agree it is a niche use case, and local refinement is usually the best next step!
Re the half-set issue, it’s an interesting point. I’m not sure whether keeping sym-expanded groups in the same half sets solves the issue - you will still run the risk of superimposing (& averaging) multiple copies of a particle in identical orientations (it just won’t manifest as spurious correlations in the FSC). Does that really matter though? I’m not 100% sure…
