In order to get the highest possible resolution, should I run local motion correction right after data cleaning (for example, removing junk particles), or after heterogeneous refinement which gives me a single class containing desired particles only? For Bayesian polishing in RELION, it seems the former is better, but I’m not sure whether this still applies for local motion correction in cryoSPARC.
Thank you so much.
Generally speaking it is better to run local motion correction directly after some basic cleaning - you want to have the maximum number of particles (of any quality) being used simultaneously for local motion correction. This does however come with a relatively large penalty computationally and in terms of disk space - most of the particle that you end up motion correcting will be thrown out.
If you are working on a protein that is biologically interesting but you already know that it’s not going to get below 2A resolution (membrane proteins, flexible proteins, etc) then I would highly recommend skipping local motion correction altogether and simply using patch-motion. It is much faster, creates much less data, does not require particle locations, and gives very accurate motion correction of both frame drift and anisotropic sample deformation. We have only ever detected a difference in final resolutions between patch-motion and local-motion with very high resolution structures (beyond 2A) and even then, the comparison is eg. 1.7A vs 1.65A.
Thank you so much for your reply.
Indeed, patch motion correction is already very powerful, and I can see that in my reconstruction. However, I’m still interested in local motion correction because an additional round of local motion correction after patch motion correction did improve the resolution (7.3 A to 7.2 A without masking, and 3.31 A to 3.29 A with masking).
In your reply, what do you mean by “particles (of any quality)”? For example, my particle is ribosome-related. Particle picking gives me ~100k particles; 2D classification removes carbon edge and gives me ~97k particles; multiple rounds of heterogeneous refinement gives me ~10k ribosome-containing particles. Which set of particles do you recommend me to run local motion correction in order to get the highest possible resolution? Also, why can the motion of “bad particles” improve the estimation of the motion track of “good particles”?
Thank you so much again for your help!