I have an error of 1.5% on the raw pixel size I used when importing the particles. I have tried the re-extraction by re-assigning particles and following directly with non-uniform refinements or local refinements. For some reason it has worked well for some of the re-extracted particles but the resolution and quality of the maps have gotten worse for many of the others.
I wanted to know opinions or views if it is really worth to invest time on what is the issue for this pipeline or if the community would use a different approach. I have read about the approaches change_apix_extract.py and change_apix_micrographs.py. Would anyone have any particular recommendation on how to proceed?
Hi, I think it’s best to redo patch CTF correction on the mics after re-importing them at the proper pixel size, then re-extract the known good particles after re-associating them to the new micrographs.
When you re-import the particles after manually adjusting the pixel size they carry incorrect defocus values because they were calculated with the wrong pixel size - this throws off the refinement, especially at resolutions better than 4 Å. Adjusting the defocus/ctf values with those .py scripts is the fastest method to linearly extrapolate what the ctf should be and avoid running patch ctf again, but this is a shortcut that gets to an approximation of what is probably correct. IMO, the truly correct way is to redo patch ctf estimation on the micrographs with correct pixel size.
@jybjybjjyb In addition to the resolution / data quality needing to be quite good for reliable Cs estimation, fitting Cs only corrects for pixel size error in computing the CTF itself. It doesn’t correct the magnification, or the dose weighting.
You really just have to repeat Patch Motion and Patch CTF. Then you can use Reassociate Particles to Micrographs and Patch CTF Extract to avoid repeating any of the other steps.
The most proper way is of course restart from beginning with the correct pixel size input. However as mentioned the difference in pixel mentioned is about 1.5% off, which means really this can only make a significant difference when we are reaching spatial frequencies where the misfitting of Cs or defocus become significant- I think this only happens somewhere below 2.5A resolution.
So if they have a 2.5A map then global CTF refinement should work and give a quick fix without the need to restart. If the map is worse than 2.5A or 3A. Then both global ctf or restart from beginning are unlikely to make a difference.
The Cs itself indeed only matters at quite high resolutions, and high resolution data is needed to fit it precisely, but pixel size error in computing the CTF potentially matters at lower resolutions because we effectively divide by tiny numbers nearby the CTF zeros to do the correction and using the wrong frequency grid moves the zeroes around relative to the Fourier transform of the experimental images.
Reusing pick locations, yes. Everything else (2D, etc) I’d redo from scratch. Unless the dataset is particularly large, it shouldn’t take that long and would provide peace of mind.