Woohoo! 
Next on the wishlist; RBMC of symmetry expanded data? 
2 Likes
Yes please! Preferably (IMO) with an option to use the full particle for alignment, then extract sym-expanded subparticles from the aligned frames:
2 Likes
Hi @rbs_sci and @olibclarke,
Is the intent with this suggest to refine subunits/subparticles that decorate the surface/edges of large assemblies and that are small and possibly flexible?
Thanks,
Kye
Yes more or less.
For example, consider a pentamer that forms part of a viral capsid, which may have some overall flexibility/squishiness. Maybe the pentamer has something bound to it that we want to classify out. Maybe the density of the pentamer is anisotropic, due to the flexibility of the capsid.
We can improve the density of the pentamer by symmetry expansion & local classification/refinement, but currently cannot practically take advantage of the improved SNR provided by either local motion correction or reference based motion correction.
For this purpose, it would be useful to be able to perform reference based motion estimation using the whole particle, and then to extract the motion corrected subparticles during RBMC/local motion correction. Motion correction of the subparticles individually may also be desirable at very high resolution, if they are of sufficient size, but I suspect in most cases just correcting for the motion of the overall particle will offer most of the benefit.
Cheers
Oli
1 Like
What @olibclarke said, basically.
Flexibility regarding how motion estimation is carried out is critical for this; whole particle is fine, although on larger objects (playing in the 250 nm+ range) will either (a) estimate motion poorly on heavily downsampled boxes (tried it) or (b) fail due to box size (also tried it), so it would be good to be able to toggle between either “whole particle” motion estimation or “unique position” (where the symmetry expanded particles overlap by more than say, half the size of the box(?), only a single particle at that position is used for motion estimation).
1 Like
Yes agreed - even in that case though, I would want to have the option to restrain the sub-particles so they are effectively refined with a single motion trajectory. The spatial prior should do this I guess, but it might be nice to have an option to force it for sub-particles, even if the overall spatial prior is somewhat looser.
1 Like
You mean on a per object basis, have the sub-particles follow the same motion trajectories? Yes, I would worry about the potential for overfitting otherwise… that said, RBMC doesn’t fit beyond FSC=0.143 the way Bayesian polishing does, or at least reports it does in the log…
Yes exactly - to force the subparticles to have the same trajectories (as they ought to do unless RBMC is effectively modeling internal flexibility/mobility of the assembly)
I can think of a case where one might want independent motion estimation on a sub-particle basis… but would want each “duplicate” to follow the same trajectory, but it’s such an edge case with so much potential for misuse if applied generally… it’s better to avoid the possibility.
1 Like
Thanks @olibclarke and @rbs_sci for your explanations!
Just to make sure were on the same page, using COVID-19 capsid/spike as an example.
Yes exactly! (Always love a David Goodsell picture 
)
But what @rbs_sci is proposing (if I understand it correctly) is slightly different - to box out the subparticles at the hyperparameter estimation stage (left side of your flowchart), but to make sure that all of the subparticles from one assembly (virus in this case) have identical trajectories, prior to extraction. This is to get around the issue that currently, RBMC cannot accommodate box sizes that are big enough for really large viruses.
2 Likes
Yupp, I see that now in his post. Thanks!
We have recorded this request!
2 Likes
@olibclarke and @rbs_sci, would sub-particle extraction from larger particle images be a sufficient option, given a large enough box size post RBMC where sub-particles could be appropriately extracted?
Maybe?
I prefer to polish/RBMC from the same size to the same size, but it might work OK. Would have to try.
1 Like