I was trying to export extracted particles from CryoSparc to Relion for a Class3D job. During the operation, should I add --swapxy and/or --reverty? The particles were motion-corrected and extracted in CryoSparc. I understand that if I want to re-extract the particles in Relion, those cards are necessary. But if I don’t need re-extraction, do I still need to keep those cards? Thanks!
If you don’t need particle locations later on (for re-extraction or polishing), then no need to keep them, but it’s handy to keep them in case you want them down the line. For particles picked/extracted in cryosparc, you need --swapxy and if motion corrected in cryosparc, possibly --inverty (not reverty), since I believe mics will be inverted after motioncor in relion relative to cryosparc. Double check your particle locations before doing a big extract or polish job. I like to run 2d classification on a small subset of re-extracted binned particles before polishing etc to make sure the locations are good.
Thanks a lot!
Btw, do I need to use star.py with --copy-micrograph-coordinates for re-extraction job? I kind of get confused with all related posts regarding this part.
I believe you only need star.py to get the coordinates and micrograph names if you originally imported the particles to cryosparc, but if you extracted them in cryosparc, you should not need it. For example, if you extract in cryosparc, then run 3D classification in relion (no star.py to get there) and import the subset to cryosparc for refinement and then want to go to back to relion, you would need star.py to get back to relion.
You would run star.py original_imported_star_file.star cryosparc_job_recent_csparc2star.star cryosparc_job_recent_csparc2star_with_coordinates.star
Always check your star file to see if the micrograph and coordinates are there. If not, try getting them with star.py
There is a bug where rln_OpticsGroupName (opticsGroup1 for default) is not generated in the optics header and you can manually add it with a text editor. Re-extraction throws an error at the very last step and you need to add this for it to work.
Any thoughts on where to look next if my co-ordinates still don’t seem quite right?
I have exported with csparc2star with all different combinations of only --swapxy, --inverty, or neither, or both. In every case 2d-classification (relion 4) of a small set of re-extracted particles from motioncorr corrected micrographs doesn’t look great.
The --swapxy only looks the best of the options, (as in there are some sensible classes), but there is still quite a large proportion of particles lost to junk. The other combinations are decidedly worse. All classes look good if I directly use a subset of the imported particle stacks directly without re-extraction, but I am wanting to try bayesian polishing.
Could the --swapxy be correct and mostly OK, but differences in motion correction between motioncorr and patch motion correction in cryosparc mean that a reasonable subset of the particles are off?
Most likely the --swapxy only is correct (that’s the usual case if motion correction is done outside CS), and there is simply a significant portion of junk particles - or your subset is too small. If you do the same small subset in cryoSPARC, or in Relion using your cryoSPARC extracted particles, does it look any different?
Instead of using a small 2D classification job, it would better to use relion_display to look at the coordinates themselves and judge if they are accurate.
PS For anyone that isn’t aware, the optics data thing was fixed quite a while back, you should update pyem.
PPS To address the actual question, it’s very unlikely there are any differences in motion correction that large. The total motion is generally on the order of a few pixels. Keep in mind motion plots are exaggerated 10x for clarity.
Thanks for the reply and explanation Daniel.
Classification of a small subset of the cryosparc-extracted particles does look much better (no junk classes), relative to the relion-extracted particles. However, I couldn’t make out anything obvious wrong from the micrographs, and committing to extract all particles refinement seems to proceed. I suspect it is may be related to the classification as you pointed out, which has always worked better in cryosparc than relion for this data (which is not brilliant).
With your clarification I think have enough info to persevere for now. Thanks very much!