Particle picking


I am working with a protein complex that has three proteins in there that make up ~150kDa complex. With this sample, I have obtained 6748 movies from GLACIOS (200 keV) microscope. Now I am at the stage of processing my cryo-EM data and I’m using cryosparc pipeline for this. I have created the workspace and I have run the patch motion and patch CTF job already. Then I did exposure curation and got rid of bad CTF fit resolution and ice thickness micrographs. This has reduced movies to 3969.
Now next step should be particle picking and my question is how should I decide if I should go for the manual particle picking or blob picker.
Also, I would like to mention that my micrograph looks very heterogeneous(attached micrograph) so it is hard for me to know the box size, particle diameter(min/max) to run the next job. And I am basically stuck at this point.

Could anyone have experience with data processing with small protein complexes like this?
Looking forward to your suggestions.Thanks a lot!

in cryosparc, particle picking is best done iteratively. a typical workflow: manual picker selection of ~100-200 particles. This also gives you the ability to see the box size around the selected particle, since “particle extraction” is part of this job type. Take the output and do 2D classification into 1 or 2 classes only, these will be templates for template picker. It’s ok if they don’t look like particles, so long as there is white in the middle of a black box. Then do template picker, estimated particle size from 100-300 angstroms, maximum peaks of probably 500-600 for images that look like this. Can do inspect picks here and slightly tailor the outcome but not necessary. Inspect picks will definitely show you each pick and allow you to “see” what various box sizes will look like (the circle around each pick changes in size with box size value). Extract with box size 200 or 256 (depending on pixel size) and run 2D classification with 100 classes. You can do another round if not satisfied, select 2-5 of these classes with 2D selection and use that as the new set of templates for template picker.

Other things: you could skip step 1 (manual pick) and do a blob picker with best-guess settings and run it for only 10 micrographs, then the first round of 2D class with 50 classes instead of 1-2 and then 2D select only the good ones for first round template pick.
Sadly, this micrograph does not look great. It’s possible that there is a high resolution structure possible from your data, but if the 2D classes don’t look good then I would be suspicious that your data is not great and need to work on sample prep and grid prep. Best of luck!

Thanks a lot for the suggestions! I just follow the steps.

Agreed it doesn’t look super promising but I wouldn’t lose all hope yet!

I would try to manually pick the rod-shaped particles (which themselves seem to be aggregating into a higher order, maybe five-fold assembly, but I’d ignore that for now) and see if you get any recognizable classes out of that (example picks below), then try to use those for template picking, or to train a Topaz model.

I would also consider whether the smaller, “dot-like” particles might be end-on views of a “rod-shaped” complex.


in which case you may want to include some of the micrographs with thick ice which were likely excluded during exposure curation…

1 Like