3D classification during helical processing in cryoSPARC?

hi, all,

how to do 3D classfication during helical processing in cryoSPARC? thanks.



Dear @niu,

I apologize for the late reply. We don’t yet support applying helical symmetry during multi-class refinement, but we’re currently re-working 3D classification and this is definitely part of the planned additions :slight_smile:

Right now, there are a few options you can try, but unfortunately none of them allow for having different helical symmetry parameters for each class. I’d first suggest running an analogous workflow to that presented in our helical processing case study up until (and including) the 2D classification stage.

  1. If the heterogeneity in the dataset is not so great as to prevent a consensus refinement from achieving sufficient detail, then I would suggest running a consensus helical refinement with the whole dataset. At this point, if you want to use 3D classification to further sort junk, or to resolve small amounts of heterogeneity, you can:

    • Run symmetry expansion with the final helical parameters shown by the consensus helical refinement job
    • Take the outputs of symmetry expansion and run them through 3D Variability and 3D Variability Display in the cluster mode. The cluster mode will perform classification in the latent space of “reaction coordinates” with a configurable number of classes
    • This workflow is explained further in part 1 and part 2 of our 3D Variability Analysis tutorial.
  2. If instead the heterogeneity is significant, and you cannot reasonably obtain a consensus refinement, then I’d recommend using ab-initio reconstruction (to get initial model(s)) and heterogeneous refinement with no symmetry applied. This is harder though, because heterogeneous refinement can’t enforce helical symmetry, and often times this is needed to get to high resolution for helical datasets (see note below).

It ultimately depends a lot on the sample, in particular, the size of the asymmetric unit. If you’re looking at a tightly packed helical filament with a very small asymmetric unit (something like tobacco mosaic virus), you will definitely need to know and apply helical symmetry, so the first method is better. If you’re instead working with a filament with a large asymmetric unit (e.g. MDA5 filament), standard cryo-EM workflows including ab-initio and heterogeneous refinement work quite well, because the symmetry is less crucial, so I’d recommend the second method.

Best regards,

thanks a lot. :handshake:

@mmclean when you use the first option you presented with symmetry expansion to 3dva to clustering, what downstream processing is still “allowed”? Since the particles are symmetry expanded I assume another helical refinement wouldn’t be, because some of the particles are not true particles. So the only thing to do after clustering would be local refinements, or to manually unsymmetrize the particles (for example Reverse symmetry expansion? [Feature request] ) and use those to do a new helical refinement?


Yes that is correct. Only further 3DVA, local refinement, and reconstruction jobs would be legal to do after expanding; if you want to do a subsequent helical refinement, you’d have to unsymmetrize. Though keep in mind, you could simply do local refinements with the expanded stack without symmetry enforced, and that would have similar output to any symmetric helical refinements.

Best, Michael