Hi, everyone. I have a question about box size. I don’t understand how to set Fourier crop to box size. If I make this smaller than extracted box size, will this affect the final resolution?
Thanks for the question. In the Extract from Micrographs jobs, the
Fourier crop to box size parameter allows you to decreasing the particles’ box size by increasing their pixel size. The final box size of the extracted particles is the value set in this parameter. Reducing the box size is typically done to speed-up downstream processing during early phases of processing a dataset, because nearly all processing will run significantly faster at smaller box sizes.
For example, if your micrographs have a raw pixel size of 0.5 Å and you instead want the extracted particles to have a pixel size of 1.0 Å, you can set the
Fourier crop to box size (pix) parameter to half the value of the
Extraction box size (pix) parameter.
By increasing the pixel size, the sampling rate of the images is decreased, and thus higher frequency information is lost during the Fourier crop operation. The effect this has on resolution depends on whether your dataset has enough SNR (and homogeneity, etc.) to reach the Nyquist resolution in the first place. If you expect a final resolution of x Å, you are safe to Fourier crop particles to a pixel size of at most (x/2) Å without adverse affects. It’s quite common to do initial processing at a smaller box size, and after an initial structure is obtained, to go back and repeat the process at a larger box size only if the Nyquist resolution limit is hit.
@mmclean I got it. Thank you very much.
I have a question about Fourier crop here:
I’m reading a paper in which they claim that their pixel size is 1.08A, and bin8 during particle extraction and got a Ab initio structure at about 12A before they bring the full resolution back. If I have a correct understanding, it is impossible. The Nyquist resolution should be 17.28A, and I don’t think the Ab initio allow them to set Initial resolution better than 17.28A.
Am I right?
Thanks and best,
That’s interesting – do you have a reference to the paper?
- Ab initio doesn’t report a resolution estimate - so there must be some more steps.
- If you use a larger box size for reconstruction than the particle box, the estimated resolution will be higher than the particle Nyquist. You can replicate this in heterogeneous refinement, for example if the particle box is 96 px and you use the default 128 px reconstruction size. This effect is a combination of correlated interpolation error and probably some legitimate “super resolution” information coming from the smoothness of the structure in Fourier space.
Thank you so much for the explanation, I learn it from your second point.
Thanks and cheers,
I’d be interested in the paper as well.