Hi everyone,
I’m refining an I1 symmetry particle and trying to resolve assymetric subunits after symmetry expansion and subparticle extraction.
My workflow is refine with I1 symmetry, symmetry expand(I1 sym), shift coordinate centers with volume align tools, reextract with smaller box size(subparticles of assymetric unit), then cryosparc’s 3D classification job(~12 million particles). The 3D class job does well with my downsampled particles and looks pretty decent but when i take the selected classes that i want and move them onto refinement(~1 million particles) jobs they turn into blobs that resemble noise. Are my subparticles just too noisy?
Please let me know if any other further info is needed and any advice would be greatly appreciated.
Best,
Tyler
Is this local refinement? Can’t do global refinement on symmetry expanded data.
Try severely restricting the angles and shift ranges; by default even Local Refinement gives too much range and the particles contributing to views “through” the capsid will drift off position and make everything a mess (block based reconstruction is something I still do exclusively in RELION partially for this reason…)
1 Like
Hi,
@tbrittai how do the heterogenous reconstructions for these classes look? Additionally, as @rbs_sci mentioned, only use local refinement and limit the search extent to something like 10 degrees and 3A shift or 5 degrees and 1A shift might be beneficial. Using the gaussian priors for pose/shift search might also be beneficial.
@rbs-sci, do you have issues attempting to do block-based reconstruction in CS? Additionally, how often are you performing this type of processing workflow? Only for large complexes/virus capsids?
Best,
Kye
My lab does a lot of work on giant viruses, so fairly frequently. Yes, normally only for larger things, although we did test it for a smaller virus with a highly flexible binding domain. Whenever I try working with giant viruses in CryoSPARC, particularly BBR, the blocks always end up very streaky “through” the capsid, regardless of the direction the capsid block is facing in relative to x/y/z. Sorry, I don’t have any current examples to share. 
2 Likes
Curious - do you see less streaking/anisotropy with Relion for these large capsids?
Yes.
It’s “easy” to track as refinement proceeds; the angular distribution plot drifts off from a nice symmetry expansion pattern - even in local refinement - to a less uniform distribution where views are dominant polar/equatorial.
Severely restricting the angular search and shift ranges help a lot, but it’s still not as clear. Having said that, I do wonder if it’s partly the convergence criteria; CryoSPARC converges a lot faster and ceases iterating angular sampling by default at 0.2 degrees in local refinement (although I’ve changed this to 0.1 and 0.05 degrees without seeing much improvement), which is an order of magnitude less fine than RELION usually converges at for whole capsid and BBR of giant viruses… I pretty consistently have the refinements in RELION converge at 0.02 degrees across multiple datasets.
2 Likes
Just quick update. Using local refinement and limiting angles and shifts helped a lot with a mask.
Unfortunately, a new issue has surfaced. After doing an initial local refinement and particles looking decently aligned I imported these particles into relion 4 using the csparc2star.py script and editing the generated star file to .mrcs . I performed masked 3D classification to get rid of non-ligand bound classes which separated out nicely. After doing a Refine3D in relion with my selected classes particles I tried importing these back into cryosparc so that I can re- extract them at a smaller pixel size. I changed the star file back to .mrc and linked the micrographs and put the star file in the particle meta path. They imported without error so I reextracted (checking flip mic in Y before extract - I was told csparc2star.py does this and needs to be reversed)and now any local refine or even reconstruction without alignment looks just like noise.
Any help or insight on this would be greatly appreciated.
Do the extract positions look sane based on where they were before?
I’ve always needed to manually tell csparc2star to inverty, so maybe turning off flip Y in extraction will make the locations correct in CS again?
Yes they look like what i would expect. I tried not flipping in Y and that was worse. Also the reconstruction only looks generally like what i would expect in the very center(I see some features of the viral trimer). Going out from the center it looks like it extracted what i wanted but its a giant smear of all different angles.
Hi @tbrittai
Would you be able to run an inspect particle picks job and look at the position of the picks on your re-imported particle stack prior to extraction?
Best,
Kye
@kstachowski In the inspect particle picks job of the import particles job, it looks like ~90% of the picks are not where they should be.
@tbrittai, did you use --swapxy or --inverty as options in pyem during conversion of CS to .star files? Or did you disable the --flipy option which is on be default?
@kstachowski i did not do any of those options.
Hi another update. I have found that when reextracting the micrographs aren’t being flipped in y even if i have the box checked. I tried updating my cryosparc to 4.5.1 and the issue has persisted. I also can’t flip my particles in y in my non symmetry expanded particles. I can however flip both symmetry expanded and non expanded particles in x during extraction. Is this a known bug or issue?
Hi @tbrittai,
This is indeed a bug, thanks for finding and reporting it. We have been able to replicate it internally and have noted it.
I took a particle stack from a project I am working on and followed the steps outlined here to import into RELION. I then performed 3D-autorefine and reimported the resulting particles back to CS (connecting the run_itrXX_data.star as the particle meta path and /path/relion/JXXX/extract/ as the particle data path), and I had no issues with re-extracting particles, homogenous reconstruction, or NU-ref. I would suggest looking at the particle locations in the .star file generated from running csparc2star and the final .star file you are reimporting back to CS to ensure that there are no large shifts in the particle locations.
Best,
Kye
@kstachowski that’s good to hear. I’ve been trying to do what you stated above and am still unable to flip my micrographs. Did you convert into a star file only using .cs and passthrough particles or did you convert with other parameters? to be clear i can import and reextract particles fine(without error) the micrographs are just not flipped in Y so my particle locations aren’t correct.
Best,
Tyler
Hi @tbrittai, I was able to do the steps I outlined above without using any specific settings in csparc2star.py or in the particle extraction job within CS (including enabling Y-flip for mics). I did convert using particles.cs and particles_passsthrouhg.csg and there were no issues with pick locations. It might be worth examining the particle locations for a few micrographs at each stage of the process to see where the pick location is becoming “messed up” (ie after csaprc2star, after classification, after 3d-autorefine, etc).