We recently collected data at National cryoEM facility. The report sent by them have initial processing from 500 micrographs with first round of 2D classes. But I am not viewing all the 2D classes given in the report, although few are similar. They mentioned the cryosparc was used for initial data processing. I tried different blob size for particle picking and also increases online EM iterations to 40 with batch size to 200,. But result did not changed. Any suggestion to catch the missing projections?
Attached images for 2D classification from national facility and our processing.
at least for the purpose of 2D class, make your box size much smaller. if it’s 200 make 140 or similar ratio.
If you are doing completely new picking from them, you cannot guarantee same results. are you also using full dataset, not 500 micrographs? again no guarantee to be same.
Run 2D again, “select” all particles which are great, obvious classes. then run 2D again on the unselected particles with the same amount of classes. 1) you may get good looking new views, 2) you will start to see bad looking new views at the bottom, which is not indicative of well-aligned particles. those aren’t views you’re missing that national resource found… they’re small representations of few junk particles that of course have to present some shape in the image.
Your 2D are much better, if you have a working model of what the protein should look like:
1)make model, 2)molmap/resample 3)import 4)create templates/2Dselect faves 5)template pick the rare views 6)2D 7) 2D select GOOD classes (not just good view but good particle alignment/2D features) 8)ab initio multiclass 9)2D good class, repeat. include topaz. If it’s a large dataset, rare views are in there, just either not favored in 2D because there are more of preferred view, or they are not picked well.
Thanks for the guidance.
Should I used the binned particles or un-binned. Binned particles 2D classes do not give distinguished features, which make difficult to sort good and bad 2D classes.
cryosparc is so fast, I have never had the need to bin particles. But this depends on your computational resources, and the amount/size of particles/data. Though, for 2D I wouldn’t expect a difference - the “high resolution features” we see in 2D are helices and strands and quaternary structure → it’s an ~8Å experiment so should be the same.
