I’ve been using 3D variability analysis in clustering mode with great success on a very tricky protein. The workflow is NU refinement, symmetry expansion with C3, then 3D variability in cluster mode. I now want to refine some of the clusters with NU refinement to get a final refinement with FSC, but how do I deal with my symmetry expanded particles? After all, they are not supposed to be used in normal refinement due to overestimation of resolution…
-Do I use NU refine on the output particles (i.e. still symmetry expanded) but in C1?
-Do I use “remove duplicates” to “reverse” the symmetry expansion? Is there a consensus on what cut off distance would be appropriate so I don’t loose neighbouring particles but only expanded particles?
-Or would I have to run 3D variability on non-expanded particles?
I searched for an answer here and in the tutorial and paper etc, but I’m still unclear on what workflow I should use. I guess the same question would apply to intermediate 3D variability mode if you output the particles.
You have a couple of options - like you say, you can remove duplicates - a cut off of half your particle diameter should be fine - and then proceed to NU, but if you have pseudo-symmetry that may lose you some of your gains from the classification you have performed using 3D-VA.
Probably the easiest though is to go directly from 3D-VA clustering to local refinement in C1 (still using the symmetry expanded particles). Local refinement using symmetry expanded particles is fine, but I would avoid global NU-refine, as this can superimpose duplicate particles, and lead to FSC funkiness.
I agree with Oli – asymmetric (C1) local refinement is best for this use case. It can be used anytime you only want alignments to be searched locally, which is exactly the case with symmetry expanded particles (to prevent superimposing the same particle twice, as Oli mentioned). Since you have C3 symmetry, you shouldn’t have to worry about the alignment search radius being too large (assuming perfectly aligned particles, the alignments would have to deviate by 120º in order to superimpose duplicated particles).
Just to add @ClaudiaKielkopf - if you have pseudo-C3, I would also try symmetry relaxation in Relion starting from a consensus C3 refinement for this data. Symmetry relaxation is not yet available in cryosparc, but hopefully will be in future.
thank you very much for your advice and input! I’m sticking with local refinement for now, I’m indeed seeing pseudo-symmetry (I should have added, it’s a very thin elongated protein and I can see in 3DV that it bends to the left and the right). I’ll look into symmetry relaxation as well at some stage, I’ve only ever considered relaxation when there is a symmetry mismatch.
Sorry to reply to such an old post, but I am currently going through the same process, and I’m not sure I understand what difference does it make that her protein had C3 symmetry and not something else.
For example, the protein I’m working with has a C9 symmetry, does this mean I should worry about the alignment search radius? What should I do about it - is there a parameter that I would have to change in the Local refinement job?
I would appreciate any insight or reference to where I could read more about this.
The search extents in local refinement parameterize the radius, in pose space, of angles that the particle is searched over. I.e., these control how far a particle is allowed to drift from its initial pose. Generally these are quite small for local refinement (often with upper bound ranging from say 2º to 15º).
If you have symmetry expanded your particles using C9 symmetry, that means the same image is now duplicated with 9 different poses (one for each ASU), each of which is spaced out by 360º/9 = 40º from each other. Ideally, after symmetry expansion, one intends for each of those 9 copies to be re-aligned by searching only over poses that cover it’s ASU, and not adjacent ASUs. That has bearing on the search radii because in principle, if the alignment radius is ~20º or larger, two particles could be re-aligned to become coincident. In practice, this is not a major concern because the job maintains that symmetry expanded particles belong to the same half-set, which keeps FSC curves valid. It is mainly a concern because it may be contrary to one’s original intent when using local refinement on sym expanded particles
Regardless, for relatively low-order symmetries, I would not worry too much about the relationship between the symmetry and the optimal search extent. If you want to be scrupulous, consider keeping it to less than (360/N)º.