Get rid of particle stacks

Dear Colleagues,

After running a Refinement job (i.e. Homogenous Refinement) we are left with the resulting 3D volume and the particle stack as output. I want to get rid of the particle stack, so I took the particles output from Homogenous Refinement and put them into an Extract from Micrographs.

At this point, I suspect that the particle stack from the initial Homogenous Refinement is not carried over after Extracting from Micrographs, then I can use them for any jobs downstream without any bias from the initial particle stack. Is this the right thinking to get rid of the particle stack?

Best,

CBEN

@CBEN, Thanks for your post. Can you please clarify what you mean by “get rid of the particle stack”?

Yes. After running a refinement job, the particles outputted from that job come with new 2D alignments. Am I right in this case?

“Particle stack” means the physical single-particle images extracted from the micrographs. You can get rid of them from disk conveniently by clearing the extract job. I prefer this to deleting the jobs because you can see what input choices you made, and also easily re-run the job if you need the particles again later. Sometimes (e.g. in Relion), these files are given a .mrcs extension to indicate a stack of multiple images. (The file format remains the same as a typical MRC file).

It sounds like you might be thinking of the particle pose information, rather than the images. In cryoSPARC, the particle .cs files from each job contain links back to the physical images, and the metadata such as refined poses. 2D classification and all 3D refinement jobs always perform new global searches without reference to previous aligned pose information. I believe the only exceptions are the local refinement job, the CTF refinement jobs, the per-particle motion job, and 3DVA.

1 Like

Hello Daniel. Thanks for that elaboration. When you say “particle pose information” are you talking about a set of particles (i.e. an output from a 3D refinement job) that has 3D alignment information? If so, this is exactly what I aim to do away with. I wish to lay out the problem I’m trying to avoid here:

(1) We have a membrane protein that’s a trimer (or at least most people think it’s a trimer) that consists of a stalk, stem, and head. An ab-initio model I generated looked very ‘trimer-like’, so I decided to pull the particles from that volume and run homo-refinement both with and without imposing C3 symmetry (2) The homo-refinement run without C3 symmetry looks good, but the homo-refinement imposing C3 symmetry looks great! So, I’m looking to see how good of a map I can get without imposing symmetry in the refinement. (3) Template picking didn’t work too well, so I decided to take the particles outputted from the homo-refinement job that imposed C3 symmetry, and do another round of 2D averaging. The 2D averages look great. Almost too good to be true, and further ab-initio models look OK too. (4) My only concern is rather or not the original particles outputted from the homo-refinement job (where C3 symmetry was imposed) has biased these 2D averages. From the sound of things, it seems that I would be OK since 2D classification performs new global searches without particle alignment bias from previous jobs, as you explained. Thanks for that explanation!

Best,

CBEN

Hi @CBEN just to confirm, @DanielAsarnow is correct - 2D Classification, Ab-initio Reconstruction, and Homogeneous Refinements all “start from scratch” in terms of pose alignment. No information from processing up that point is used, other than the initial model provided for refinement, and the raw particle images and CTFs. So the 2D classification you are seeing is fully “unbiased” in the sense that the symmetry imposed upstream has no effect.