While testing CryoSparc with our particles (500px box size), I ran into the following problem. After running two rounds of 2D classification, I used the resulting particles as the input to Local motion correction with a 1250px extraction size as per recommendation. However, the resulting particles cause problems in 2D classification and Homogenous refinement, as the increased box size seems to exceed the maximum possible with our NVIDIA Quadro P5000 GPU (16GB memory).
Is there a way to simply crop out the size required for our particles? What is the recommended workaround in this situation? Downsampling works, but results in a loss of resolution.
The only similar article I found was based on version 0.3.9, so I’m curious to know whether any updates.
How large is your actual particle? A 1250px box size for local motion correction could be reasonable for a very large virus or similar, but otherwise may be overkill. There are two options - you can either re-do the local motion correction with a smaller box size (and in fact, going below the recommended 2x particle size will not necessarily ruin the result with a large particle), or as you already mentioned, you can downsample the particles with the “Downsample” job type, if you do not expect to reach a resolution that is better than the Nyquist rate after downsampling.
Re-doing local motion correction is probably the best bet. If the particle is really large enough and goes to a high enough resolution to warrant a 1250px box size, then there isn’t really a way around the large memory requirement.
We are hoping to work on a CPU version of some job types or a memory-optimized GPU version that can handle very large boxes, but it is not on the short-term roadmap.
Thanks for your reply! Our particle is about 500px in diameter. I will retry the local motion correction with a 900px box size, as this is the largest that seems to work.