CtfRefine in Relion3.1 after Cryosparcv2 homogenous refinement


How would I go about running CtfRefine in Relion3.1 after CryosparcV2 homogenous refinement? I need a particles star file and a postprocess star file to run . I used pyem command below to get a particles star file. Could I use the half maps from cryosparc job and the converted particles.star file to get a postprocess.star file directly?

python csparc2star.py - --copy-micrograph-coordinates particles-that-was-used-for-import-into-cryosparc.star cryosparc_P5_J71_particles.cs cryosparc_P5_J71_particles.star

1 Like

I saw the thread

mentioned _rlnRandomSubset being random and not saved in cryosparc . So I would have to refine again in relion or run local refinement in cryosparc? And that halfmap could be used in relion postprocess. If I am using the same mask from the homogenous run would I need to extract any differently?

The half map split is saved by cryoSPARC after a refinement, and rlnRandomSubset should be populated when you convert back to a .star file. The other thread is talking about what happens when you import particle alignments or re-run new refinements in cryoSPARC.

How would you suggest extracting and preparing my particles.star and postprocess.star after cryosparc when my end goal is to merge datasets and run CtfRefine?
I get a very noisy map after pp if I use the half map in relion to get a postprocess.star file.

It looks badly oversharpened. What is the expected resolution, and what does the map look like when you simply reconstruct the parameters converted from cryoSPARC?

The expected resolution is 4A. When I ran relion_reconstruct using cryosparc parameters the map looks like the cryosparc result. Same for the half-maps.

Is the postprocess map you showed masked or unmasked? It’s normal for unmasked sharpened maps to have a lot of dust, since you’ve just boosted the high-frequency components a lot. You might want to try some manual b-factor selection, like -80, -100 instead of the automatic one which might be a lot higher.

You can also see if it works better with relion_reconstruct --subset 1 to do two half-map reconstructions instead of the ones from cryoSPARC. Another option is to do one iteration (or just let it converge) of local angle search in Relion.

It was unmasked. Good to know its normal. Thank you! I will try what you’ve suggested.