Hi all,
I have a question about the recently released “Reference Based Motion Correction” recently released in the nice last update.
Based on the fact that the motion correction depends on a spatial prior to allowing a better motion estimation despite the low SNR, To what extent a low density of particles by micrograph will influence the robustness of the estimation? For example, if a micrograph contains 2 particles from the final subset I guess the spatial prior will not help much. If you deem that low particle density is redhibitory for a good motion estimation what would be the limit of particles by micrographs to get a good motion estimation?
Best,
Julien
Hi @Jubous,
Great question. Anecdotally, I’ve seen reference motion succeed on very sparse datasets, despite the fact that in principle you’re right: the spatial prior has less data to work with on sparse micrographs. It seems that the temporal prior can perform surprisingly well on its own. If we get a large volume of feedback indicating that there really is a servere problem, we might implement some sort of warning in the future. At the moment, we only prohibit micrographs with a single pick.
–Harris
2 Likes
It would be nice to be able to override the single-pick limit, as with some of the samples I work with are lucky if we acquire four or five particles on a micrograph (the fun of working on giant viruses!) so after 2D classification and 3D heterogeneous refinement/classification the possibility of a single particle per micrograph is very real…
Hi @rbs_sci, thanks for the feature request. We’ll consider this for a future release.
Are there any news on this feature (being able to use movies with single particles)? I’m having a similar issue. I have a mixture of two similar complexes in a 80/20 ratio. I performed separate reference based motion correction jobs on the two particles sets and ended up loosing about 15% from my already small dataset. Or is it better to join both particle groups and perform reference based motion correction on the unified set?
Yuval.
You can feed RBMC two input sets (or more) so for this case I’d go with that.