Local refinement with symmetry expanded particles

Hi, this is a more theoretical question because I’m not sure I understand.

I expanded my particles in C2 (so I understand I have now double the amount of particles) and then ran a local refinement using a mask from the previous homogenous refinement, so covering all the dimer, not only one monomer. Is this correct to do? The FSC curves and map look OK but I’m not sure if this correct or I should only run a local refinement with a mask in one of the monomers?

Thanks in advance.

I must add:

  • The resolution improved only slightly, but there is a sharp straight drop and then little bump on the corrected FSC (not sure what this means).
  • The map quality has greatly improved.

Hi @LTP,

This is definitely an intended workflow for symmetry expansion. In general, symmetry expansion + local refinement (using any mask – could cover the whole dimer or just a monomer) can help whenever you are interested in seeing any features that would have been averaged together if symmetry was rigidly enforced. The only real “catch” with this workflow is that you should keep the GS half-set splits the same (i.e., make sure the “Force re-do GS split” parameter in local refinement off, which it should be by default). This ensures that all of the expanded copies of one particle stay in their same half-set, meaning you don’t have to worry about duplicate particles.

If you are seeing a quality improvement using local refinement with the mask covering the whole dimer, it’s possible that this is due to the finer angular search that local refinement does. I’ve sometimes seen FSC improvements on the order of 0.1 - 0.15 Å using symmetry expansion and local refinement (relative to just standard symmetric refinement) that I believe could be due to this – although this was with a relatively flexible helical filament. If you have non-uniform regularization & marginalization enabled, it’s likely also due to that.

Usually, if you are explicitly interested in resolving inter-subunit flexibility, you can try a mask covering just the monomer, but the success of this depends a lot on the align-ability of the monomer.


Hi @mmclean, thanks for the answer. One more question related to local refinement with or without signal subtraction: does it matter if the data is binned? I am working with 2x binned particles (pixel size around 2.5A instead of 1.25A) and because I’m far from Nyquist, I’m doing all refinements still binned 2x. Sometimes in local refinement jobs I get weird FSC curves that state the resolution is 4.9A (the curve never reaches nor goes near zero, just stays very high), which is Nyquist. Should I work at full resolution even if the resolution in my reconstruction doesn’t extend further than Nyquist at 2X?

Thanks for all your help!

Hey @LTP,

Both subtraction and local refinement should work as usual with downsampled data. If you’re seeing all of the FSC curves (even the unmasked FSC) quite high all the way out to the max resolution, it could likely be a genuine resolution increase, in which case yes, moving to the unbinned data could bump up the resolution more. You would have to re-do signal subtraction though – but you can use the same alignments from the unbinned data by overriding the “blob” input using the job builder’s low-level results feature.

The other main causes of unexpectedly high FSC curves could be:

  • Duplicated particles crossing the GS split, which you do have to be careful of when working with symmetry expanded particles (from above).
  • Using binary / not-soft masks for any refinements – this often makes the loose and tight masked FSC curves wacky (see Oli’s post where we figured out that the cause of his strange FSCs here was the binary mask)