Re-extract particles after reference based motion correction

Motion Correction
1

I used reference based motion correction on a set of particles(C5 symmetry) and it works well(the resolution increased from 3.30A to 3.10A ). Then I try to extract the sub-region of this particle(symmetry expansion for the original particle------recenter to the sub region------re-extract the sub-region with a smaller box-----use homo reconstruct only to check the result), the resolution of the reconstruction drops to 5 A. However, when I follow the same pipline to process the particles without RBMC, the resolution of homo reconstruct only is 3.54 A(the resolution of refined whole particles is 3.30A). So I would like to know why reextracting the particles of RBMC leads to a decrease in reconstruction resolution and what improvements should be made?

2

If RBMC only refine the particle image, maybe I should crop the sub particle from the particle image instead of re-extract them from the micrographs?

3

See discussion here: Reference-based motion correction with symmetry-expansion

Basically at present cropping from the RBMC-corrected particle image is the only way to do it. Just make sure that the box you used for RBMC is big enough to accommodate the box of the subparticle.

Cheers
Oli

4

Hi @Reborn,

RBMC can modify the particle centers, which is why homo reconstruct after RBMC will be much worse than another refinement. We always recommend re-refining after RBMC for this reason; another refinement will produce new poses and shifts which are correct even if the particle “drifted” during RBMC. So ultimately, yes as per Oli’s post, cropping from the RBMC corrected particles is probably the only way to accomplish your aim at present.

– Harris

1 Like
5

Thinking about it, this adds one more wrinkle - if you want to crop around the subparticle centers, then you probably need to re-refine the whole particle first (local refinement should be fine), as otherwise the extraction centers of the subparticles will be wrong, due to drift of the entire particle.

Also can confirm re needing to re-refine - in a recent case we went from 2.34 (pre-RBMC) → 2.93 (reconstruct after RBMC) → 2.25 (local refinement after RBMC). I guess the drift would mostly manifest as errors in offsets, so maybe local refinement with small orientation searches & larger translational searches might do the trick?

1 Like