How does box size translate to different jobs?

Hi,

I have a long particle so am trying to use a large box size to resolve more of the structure. I have re-extracted my particles with a box size of 588 and fourier crop box of 388 (raw pixel 0.72). I wanted to first run a few heterogeneous refinement jobs for 3D classification before running homo and NU refinements, however my GPU runs out of memory and the hetero refinement job fails with the 588 box size. The job only goes to completion with a boxsize of 400.

My question is, when I perform further refinements on the output particles from this hetero refinement job, will these be with the 400 or 588 box size?

Thanks,
Anokhi

1 Like

Am I understanding correctly

  1. Input particles for the failed job were extracted with a 588 px box size without Fourier cropping.
  2. 588 was explicitly specified in the “Refinement box size (Voxels)” field of the heterogeneous refinement job.
  1. 400 was explicitly specified in the “Refinement box size (Voxels)” field of the heterogeneous refinement job.
  2. At what “Extraction box size” and “Fourier crop to box size” were particles extracted for this job?

@wtempel

This is how I extracted the micrographs. The raw pixel is 0.72 A. I wanted to set Nyquist to 2.5 A.

Yes the input particles fail with the 588 box size specified in the refinement box size (no Fourier cropping).

This is how the heterogeneous refinement works. If the refinement box size (voxels) is set to 588 it doesn’t run to completion.

Your box size after Fourier cropping is 338. Why do you want to run hetero refinement in a box size of 588?

The output particles will have the same size as input - 338 - regardless of what box size is used on the fly for calculations during the job.

Cheers
Oli

Hi @olibclarke,

I didn’t realise that the box size would only be 338. I thought that the Fourier cropping was binning the particles to a lower resolution (I followed this post Fourier crop to box size - #2 by mmclean).

I essentially want to have a large box size (due to the long particle), but instead of using the raw pixel size of 0.72 A, I just wanted to set it to 2.5 A to make the processing faster (or even possible due to the GPU limitations). Did I go about doing this incorrectly?

My calculation for the Fourier cropping parameter was (0.72A/(2.5A/2))*588

Hi @AnokhiShah, the Fourier cropping is binning the particles to a lower spatial resolution - effectively resampling them to a smaller box. In your case, cropping from 588 to 388, with an original pixel sizeof 0.72, that would equate to a pixel size of ((588/388)*0.72)=1.09 Ă… (and a Nyquist resolution of 2.18Ă…). Does that make sense?

Hi @olibclarke, that makes complete sense and is definitely not what I wanted to achieve. Is there a way to bin by pixel size rather than spacial resolution in that case?

These are the same thing, effectively - altering the pixel size by Fourier cropping alters the maximum achievable resolution. But it’s not really a problem - I would work with binned particles (usually binned to ~4Å/pixel initially), for initial rounds of clean up and classification, then re-extract at either full scale or 2x binning depending on the expected resolution once a “good” set of particles has been identified.

Cheers
Oli

1 Like