Cropping subtracted particles after recentering

[miwoodso]

I have a psuedo-C6 complex with many flexible sub assemblies. I’ve had some success with local refinement after symmetry expansion and signal subtraction to isolate individual sub-assemblies, but to continue (e.g. with flex training) I’d like to recenter and clip them out to a smaller box size. Several older topics (Volume alignment tool - #8 by nameless_wonder) suggest a workflow to do this:
symmetry expand
subtract
volume tool: recenter to mask
downsample: crop
However, when I run a homogeneous refine on the cropped particles, the results are a mess.

Local refinement before recentering gives 3.43 A while homorefine after cropping gives 5.96 A, and this is generous. The problem seems to be the downsample job, a reconstruct only job using the recentered particles gives results similar to that of the full pseudo-C6 complex.
Any thoughts on what would cause this? I have tried with and without specifying a Fourier crop box size equal to the real-space crop size. Original boxes are 512 px, I have tried a few crop sizes around 200 px.
Cryosparc version is 4.6.2

[wtempel]

Thanks for this question

@miwoodso

Cryosparc version is 4.2.6

@miwoodso Did you switch 2 and 6 by accident, and is the true version 4.6.2?

[miwoodso]

Yes, that’s correct. I even went back to double check and still got it wrong somehow. I’ll edit the original post to reflect this.

[rposert]

Hi @miwoodso, that map certainly looks like a mess!

Can you tell us a bit more about what you’re trying to do here? We might be able to figure out a better workflow for you if so.

1. Can you share your particle size (in Å) and your pixel size (for the original, 512 px boxes)?
2. Are you subtracting all but one of the asymmetric units from the symmetry-expanded particles? Essentially leaving only one ASU in each image?
3. When you say “downsample: crop”, can you tell me a bit more what you mean? Are you downsampling and cropping during a particle extraction? If so, what values are you setting these parameters to, and why?
4. Can you tell me a bit more about the Homogeneous Refinement you showed an image of? Were these the cropped particles and cropped volume?
5. Can you share the 3.43 Å result from Local Refinement before cropping?
6. Does a Homogeneous Reconstruct Only job using the cropped particles look reasonable/as expected?

[miwoodso]

1. About 190 Å long, pixel size is 1.362 for all boxes
2. I’m subtracting all of 5 ASUs and all but one complex from the 6th ASU. Each ASU has ~4 components, one rigid and 3 that move independently.
3. I’m using a “Downsample Particles” job, which is in the ‘Extraction’ group in the Builder tab. I am not actually downsampling the particles, only cropping. I don’t know another way to crop particles. I’m setting the “crop/pad to box size (pix)” parameter to 208. The particles themselves are less than 140 pix along their longest axis, I thought 34 pix extra on each side would be enough. I tried a tighter box originally, but when I got bad results from local refinement, I tried expanding it but that did not help.
4. Yes, these are the cropped particles. The input volume is not cropped, though. I’ll try running it again with the results from reconstruction only as the input volume and edit the results in here. EDIT: this did not affect the results
5. here are the real space slices:
   
   
   
   you can’t see much because the unmasked area is relatively small. Here is a screenshot of the map in chimera:
   
   
   
   This local refinement job is before recentering as well as before cropping. I have tried reconstruction only after recentering but before cropping, and the results are similar to what I’ve shown here (but worse due to uncorrected local motion/flexibility)
6. No, it looked bad, but not nearly as much as the homogeneous refinement in the OP. It is “smeared out” along a vector that is not one of the coordinate axes:
   
   
   
   
   

[rposert]

Thanks for that information @miwoodso.

Let’s try to pin down exactly where the problem is occurring. Could you try the following workflow, and share images (from ChimeraX) of the maps made from each Homogeneous Reconstruction?

1. Symmetry expand the particles in C6 and perform Homogeneous Reconstruction
2. Subtract 5 of the 6 ASUs, but leave the entirety of the 6th. Perform a Homogeneous Reconstruction of the subtrated particles.
3. Use Volume Alignment Tools on the result of step 2 to recenter the volume and particles on the remaining ASU. Perform a Homogeneous Reconstruction.
4. Use Downsample Particles to crop the particles to the desired final size. Be sure that Recenter using aligned shifts is turned on. Perform a Homogeneous Reconstruction.

This will help us determine where along the pipeline you describe the volume degrades.

Thank you!

[miwoodso]

I strongly suspect the bolded portion of your point 4 is the problem. I’ve re-cropped the particles with recenter using shifts turned on. Both reconstruct only and homogeneous refinement look much better. The refinement still looks substantially worse than the uncropped local refinement, likely because I didn’t supply a mask. I will try that and report back.

[miwoodso]

I’m not able to substantially improve the homogeneous refinement; tweaking the mask, limiting alignment resolution, or other options don’t lead to a map as good as the local refinement of the uncropped particles.
However, local refinement of the cropped particles gives the best resolution so far (3.20 Å) so I believe my problems with homogeneous refinement of the subvolume are unrelated to the original problem of this question, which has been answered. I’ll ask about the discrepancy between homogeneous vs. local refinements in a separate question.