I have a data set collected on Arctica with SerialEM in 3 x 3 pattern. I have separated cleaned up /refined particle stack into 9 exposure groups (there are ~30K particles per group). Running Global CTF Refinement significantly alters Spherical Aberration for 5 out of 9 groups: G1 = 2.706, G2 = 37.6, G3 = -43.4, G4 = -24.7, G5 = -56.5, G6 = 2.7, G7 = - 38.92, G8 = 2.71, G9 = 2.67 mm.
Is the extreme SA value a some sort of local minimum? Is there anything we can do to limit the search range (or is this data just particularly awful)? For this job I have used default settings with 5 iterations. The large SA value is estimated on the 1st iteration and changes only slightly with later runs. Fit resolution range used was 10 A for Min and 8.7A was estimated for Maximum. I am currently re-running several groups and varying fit resolution values. Version 2.12.4.
With fit going only from 10A to 8.7A (the 8.7A is estimated from FSC), there really isn’t enough signal to reliably estimate high order terms like spherical abberation and tetrafoil. Beam tilt may still be possible to some degree. I would change the settings of global CTF refinement to:
Fit Tilt: True
Fit Trefoil: False
Fit Spherical Ab: False
Fit Tetrafoil: False
Also, in global CTF refinement, iterations > 1 generally don’t do anything since the alignments of particles are not changing, and the odd and even CTF terms are independent of each other. The iterations parameter was actually added because of upcoming support for anisotropic magnification estimation (in which case, repeating CTF refinement does help).
Generally we’ve seen improvements from global CTF refinement only with data that already reaches 4 or 3 A.
Just want to follow up on this question. Recently I also had a case in which I turned on both Optimize per-particle defocus and Optimize per-group CTF params in Homogenous Refinement, and the spherical aberration changed from 2.7 to 3.4. The final resolution reached 3.4A. And based on our fitted model everything seems to be reasonable.
Because the dataset was collected on a Krios, I would imagine such a big change will be unlikely. So I repeated the refinement but turned off both Optimize per-particle defocus and Optimize per-group CTF params, and the resolution is also ~3.3A and everything is basically the same. Can you comment on this type of behavior? Please let me know if you need more information, such as job logs, etc. Thank you very much for your support.
Actually, @DanielAsarnow mentioned in another thread (How to refine pixel size?) that this type of FSC looks like a membrane protein dip. Can you elaborate more on that or suggest some references for reading?
Back to the spherical aberration question, on that website (https://www3.mrc-lmb.cam.ac.uk/relion//index.php?title=Pixel_size_issues) it was also mentioned that SphericalAberration will remain the same while the error in Cs is expressed in EvenZernike. Does cryosparc also has a similar parameter? For example here is my final refined parameters.
[CPU: 10.21 GB] Phase shift (rad): -0.0299637569282
We have also seen refinement of Cs being unstable at moderate resolution (~3-3.5Å), in instances where we know the pixel size is accurate. Usually in such a case we just fit tilt/trefoil and do not attempt to fit Cs/tetrafoil, which usually don’t make much of a difference anyway at this resolution unless aberrations are severe.
I agree with Oli about the resolution and tetrafoil/Cs refinement.
@hwangab how did you calibrate pixel size? Diffraction spots aren’t that accurate; you really need to carefully compare a < 2.0 Å apoferritin map to the crystal structure, or use refined Cs values from such a structure. (And these two methods should converge).
It’s an electron microscope and Cs is measured in milimeters so it really can’t be very different than the manufacturer value. We just got a microscope with a custom low-Cs objective and the Cs refined value for high res apof maps is almost exactly the design value.
Just to add to this, @apunjani, it would be helpful if there was a way to reset per-group CTF parameters in either jobs where they are refined on the fly, or in standalone global CTF jobs. E.g. if I realize I have been refining with higher order CTF switched on, and Cs has refined to a high value, I would like to be able to run another refinement where the input Cs has been reset to the default input value (e.g. 2.7mm).