Any ways to randomly phased the models in CryoSPARC 3.2

[Linlin]

Hi,

I want to remove junk particles as many as possible after 3 rounds of heterogeneous refinement with a single good model and several bad models. I noticed that “random-phase” 3D classification in relion can be performed to remove particles of interest without high resolution information. random-phase” means to randomize the phase information of a good model in Fourier space beyond resolution to get a bad model, for example my final resolution of a good map is 3.8 angstrom, I can randomize the pixels between 10 and 3.5 angstrom. The two models can be used for next round of heterogeneous refinement. So Any ways to randomly phased the good model in CryoSPARC 3.2 to obtain a bad one?

Thanks.

[olibclarke]

Not at present although it has been suggested in the past, and I agree it would be great to have, see discussion here:

Removing damaged/bad particles by multireference refinement with low pass filtered references?

That would be most helpful, although the “manual” precedure seems to work surprisingly well, at least for this dataset. The difference is that random phase is introduced only once at the beginning of heterogenous classification (doing it at each cycle requires modification of the code, of course), but it still works. I can share the dataset, if you like. Again, this is anecdotal, but in our case Relion seems to do less well than cryosparc at removing junk particles. (BTW, as in the original post…

[olibclarke]

I haven’t tried this, but you could probably fake it in heterogeneous refinement by using two references, both from the consensus refinement but one filtered using a lowpass filter of a specified resolution.

Then turn off the random assignment iterations, and change the initial resolution to be closer to the higher resolution of the two references. Switch off the resolution convergence criteria (so the only convergence criteria is particle assignment), and set a very large batch number so it uses the whole dataset at each iteration, and that should be about as close as you’ll get to the random phase classification procedure in the current version of cryosparc (and then iterate this procedure with progressively less aggressively fitered “junk” references). Hope that helps!

Cheers
Oli

[olibclarke]

I have tried this workaround on a test dataset and it seems to work pretty effectively. You probably also want to switch off “Use max FSC over classes for filtering” - in this case we want the smaller, low resolution class to be more aggressively filtered, so it doesn’t converge to high resolution