@apunjani @mbrubake It would be useful from a data collection perspective to know whether more particles will make your map better. A cryoSPARC experiment that compares map resolution versus number of particles included in the map would allow the experimentalist to make this decision. Several people have proposed other methods of estimating this return on investment (e.g. ResLog).
@JohnRubinstein great idea I looked up ResLog, seems like a perfect place to start.
I believe the ResLog type plot can be computed with only a single pass through the data, since we just need to compute FSCs repeatedly as images are accumulated into the M-step volumes. So it could be quite fast.
Perhaps at first we can keep it a separate job type from refinement, then if it’s lightweight enough we can bake it right in.
This sounds great and would be very helpful! I wonder if there is anything one could do in a similar vein to identify an optimal set of micrographs? Micrograph screening manually is still very tedious, and apart from removing the obvious junk, it’s often unclear how stringent to be. I wonder if cryosparc could help to measure this? E.g. by looking at impact of adding particles from a single micrograph to a minimal reconstruction with a random set of particles, normalizing somehow for the number of particles added?