Our workstation has a 2TB SSD installed along with 4 GPUs. Recently, we frequently run into problems of SSD not having enough space when multiple people are processing their data at once. This is partially because we are loading all of our particle stacks to the SSD cache, even when we’re only using small fractions of it. Currently there’s no way of extracting particle subsets with local motion corrected so we can’t just load a subset of particles to the cache. It would be helpful if we could correct local motion using exist trajectories in particle extraction job so we don’t need to re-run local motion correction again for subset of particles.
We also tried to work around this with the “Downsample” or “Particle subtraction” job but the result are a bit worse compared with original paticles.
This is a good idea, but the downsample particles job should do what you want - if you just give it the subset of particles (from local motion correction) and set the downsample size the same as the input size, it should write out compacted .mrc files with only the particles you want, but they will be identical to the input particles. The downsample particles job doesnt/shouldnt change the particle data at all in this case. How did you determine that the results were a bit worse?
Thanks for replying! I did the downsample job without putting any number for the box size, assuming everything will be the same, but when I re-run the refinement but it got to much worse resolution. 4.1 A vs 5.6 A to be exact. Not sure what’s causing that. I actually did the same for different classes and they all looked much worse.
I checked the code and it does seem like it should do nothing in the case where you lave the box size blank… so this is bizarre!! We’ll continue looking into it and try to reproduce here
Hi @Hui, we made a fix to the Downsample Particles job in the latest cryoSPARC v2.15, released yesterday. It should no longer worsen resolution. Let us know if you run into any more issues with it!