Best practices (December 2020) for finding high res subsets


I absolutely LOVE cryoSPARC for all aspects of data processing EXCEPT 3D classification. My current workflow is to queue up like 5 different particle picking parameters, 2D classify ad nauseam, combine datasets, and remove duplicate particles after I find stable 2D classes. However, from here all I really have to classify the data further is ab initio reconstruction. This is OK, but IMHO it (1) isn’t amazing at finding similar subsets of particles with slightly different conformations and (2) isn’t amazing at finding a “magic” subset of particles that refines to a higher resolution than the consensus refinement. A common practice in Relion is to use 3D classification without alignment to find subsets of particles that refine to higher resolution than the consensus refinement. Additionally, in Relion I can signal subtract and recenter on a specific x,y,z coordinate, giving me really amazing control over focused classifications. However, the bookkeeping to go back and forth between the two programs is very arduous.

So…what are the tips and tricks to get the most out of cryoSPARC? Do people still go in and out of Relion? Can I leverage functions in cryoSPARC to act more like a traditional 3D classification?

Any input is helpful!

1 Like

To be clear, I’m specifically looking for strategies to sort particles in 3D, not to decode variability using 3DVA or Local Refinement.

We use several rounds of heterogenous refinement to achieve subset classification. It is also much faster than ab-initio.

1 Like

Great! I’ll revisit this. Tried it way back when, but I’ll give it another go.

In our hands the best results are

  1. Starting with a smaller box size (~6A max res)
  2. Using the output volumes from prior runs as input to the next het refine
  3. Increasing the box size (~3A res) and the batch size (1000 to 3000) in subsequent runs as the het res approaches the res limit.

Probably different for each case but this has worked very well for us.