Hi,
I’ve noticed that when spherical aberration, tetrafoil and anisomag are refined in global CTF (at least when run on-the-fly in homogenous refinement), a small non-zero phase shift is refined, for data where the phase shift ought to be zero (collected without a phase shift, and where no phase shift was refined during Patch CTF).
Is this expected, and what is the explanation? Sorry if I am being dim and missing something obvious…
Cheers
Oli
@olibclarke Please can you post the output of the command (replacing P99, J199 with the job’s project and job IDs, respectively):
cryosparcm cli "get_job('P99', 'J199', 'job_type', 'version', 'params_spec')"
Hi Wolfram, here you go (not the same job, but same issue/observation):
{'_id': '687e27b9341f26de60df8d53', 'job_type': 'ctf_refine_global', 'params_spec': {'crg_do_spherical': {'value': True}, 'crg_do_tetrafoil': {'value': True}, 'crg_num_iters': {'value': 2}}, 'project_uid': 'P11', 'uid': 'J371', 'version': 'v4.7.1'}
Dear @olibclarke,
When fitting any of the even order terms in the CTF (spherical aberration, tetrafoil) and anisomag (due to induced updates in defocus), all other even order terms are re-fit, and this includes the “phase shift” which is a constant (0th order) term in the CTF phase.
However, the “phase shift” encapsulates both a phase-plate induced phase shift, as well as a small correction arising straight from the amplitude contrast. This value is default set to 10% in Patch CTF, a small value, but from my understanding the typical recommendation value of between 7-10% is imprecise and can actually vary. So in effect, re-estimating the phase shift really is more of a correction on the amplitude contrast.
(The phase shift is the w term in Singer and Sigworth, 2020. Eqn 8 shows a simplified CTF model omitting astigmatism, 3rd order terms, and tetrafoil)
Best,
Michael
Thank you Michael - I figured it might be a fudge factor for AC, but couldn’t find it explicitly stated anywhere in the docs/guide that this was refined in the general CTF model in CS, so this is helpful to know.
Out of curiosity, did the grid have a carbon film support?
It did as a matter of fact! Thin carbon
Hi @mmclean and @olibclarke, I found this discussion revalent as I am processing a Volta phase plate aporerritin dataset and have the same global CTF refinement settings. @mmclean if I understand it correctly, the phase shift is refined globally as a 0th order term in CTF for an exposure group, does it mean this value is not refined per micrograph? I want to understand, to what level, is phase shift being refined, and what about phase shift refinement in per particle defocus?
Thanks!
Yue
to what level, is phase shift being refined, and what about phase shift refinement in per particle defocus?
The only jobs that modify exposures’ phase shift values are the CTF Estimation jobs (specifically, Patch CTF Estimation if movies were imported as phase plate data, or if Do phase refine only is enabled). Patch CTF can find a different phase shift value for every micrograph – it’s not restricted to finding a constant phase shift across the entire Exposure Group (EG).
In Local CTF Refinement, particles’ phase shift is not refined at all. The only jobs that modify particles’ phase shift values are Global CTF Refinement.
the phase shift is refined globally as a 0th order term in CTF for an exposure group, does it mean this value is not refined per micrograph?
In Global CTF Refinement, phase shift is only refined if Fit Spherical Aberration is enabled. The reason for this is due to the interaction between CTF parameters when attempting to estimate one of them. The set of Defoci, Astigmatism, Spherical Aberration, Tetrafoil, and Phase Shift comprise all of the CTF basis functions with even order/parity. Misspecified higher-order CTF parameters are able to induce incorrect compensations in lower-order CTF parameters of the same parity. For example, misspecified spherical aberration coefficients (order 4) can introduce systematic biases in the defoci (order 2). Likewise, another example is discussed in more detail by Zivanov et al, 2020: if the amount of coma/beam-tilt (order 3) is unknown, it can actually introduce systematic biases in particle shifts (order 1).
In order to fully correct for the effect of these high-order terms, it’s necessary to estimate an additive correction term to the lower order CTF parameters of the same parity. The CTF parameters of a given parity are tangled in a sense. Thus, if spherical aberration refinement is enabled, the job will seek an additive correction to the defoci and phase shift that is constant across the exposure group.
Note that if data was imported as phase plate data, it will not typically be the case that the starting phase shift values are constant across all of the micrographs in the exposure group as mentioned above. For example, say find that with our new spherical aberration value, all micrographs in the exposure group should have their phase shift reduced by 0.1 radians. We therefore subtract 0.1 from each micrograph’s existing phase shift estimate. So the value we found is a global value (i.e., all micrographs in the exposure group are shifted by the same value), but they will still have per-micrograph phase shift estimates.
Thanks for bearing with my response as it’s quite a complicated/unintuitive topic – definitely let us know if you have any more questions!
Hope this helps,
Michael
thanks Michael! That answered all the questions I have!
