Refine pixel size with fixed Cs?

Hi, would it be possible to add to Global CTF refinement an option to refine the pixel size, while keeping Cs fixed? Often Cs is known pretty precisely, and at high resolution can be useful to estimate the true pixel size.



This feature would be very useful for me as well. I just ran into a situation where I wound up doing this manually (calculate difference between Cstrue and Csapparent, change pixel size, refine Cs, recalculate, re-refine, etc) and an automated process would be very helpful.


Yes we had a similar situation! It would also make for easy/quick calibration at different mags, using a high res apoferritin sample, for example


I equally support this feature request!

It will be extremely relevant to almost any EM user as we more and more commonly go to high-resolution and appreciate calculating precise distances and comparisons between our models submitted cryo-EM derived models.

Greatly appreciated! =D


Would be useful for me also.


Hi all,

Thank you for the feature request. We have triaged estimation of the pixel size (with fixed spherical aberration) as an addition to our CTF refinement jobs.

For those who do this manually, do you follow a procedure similar to that described in RELION’s Pixel size issues (How can I estimate the absolute pixel size of a map?) documentation?



I’ve been calculating the pixel size as described there (Nominal * (CsTrue/CsApparent)^(1/4)), with CsTrue being the listed value of the microscope, and CsApparent being the refined value from the Group CTF refinement. Then I get my new pixel size, re-refine Cs with that pixel size, re-calculate the pixel size from the new CsApparent, etc.

It’s obviously not perfect because technically one of the variables has to be known and constant. But it works well for if pixel size is grossly off, which is usually apparent when looking at the high-resolution portion of the CTF fit, or if the Cs refinement starts to take off and go very far from published values. In my case, at 36,000x with a published Cs of 2.7, it was refining to 3.3-3.4, which worked out to be a 4% pixel size difference. Now it refines Cs to 2.68 instead, after going through a few iterations of that process. In one 1500 micrograph dataset that had 200 micrographs at > 3Å or better CTF fit, after pixel size refinement, it jumped to 1200 micrographs > 3Å CTF fit.