Defocus refinement/local CTF refinement



Hej all,

I’m mocking around with defocus refinement in my reconstruction of a small protein complex (~100kDa). The defocus range histogram is a bit wider than the one shown in the tutorial but I’m not so much worried about that. What strikes me is that the defocus error landscape looks very different from the tutorial, by far not such a beautiful local minimum.

Does anyone else get these error landscapes? Does this mean the defocus refine is bad? Is this because the refinement itself isn’t great (which it isn’t, probably sitting around 6-8A, it’s a challenging reconstruction)? I have additional tilt data from this particle which is why I’d really like to refine the per particle defocus…

These are the parameters I used. I tried using a different mininum fit res in other refinement jobs which din’t seem to help, I haven’t tried it on this exact one actually… Params

Any help is much appreciated!
Thank you,


Hi Claudia,
Defocus refinement is not usually helpful at resolutions below ~3-4 A, unless the initial CTFs were mis-estimated or the grid was tilted during collection. The refinement uses projection matching just like aligning the particle angles, the projections need to have good features of about the resolution where small tweaks to the CTF values are meaningful.


Hej Daniel,

thank you for your fast response! Right, okay, that makes sense. For now, I’ll only refine my tilted particles then and see where that leads me.



Hi Claudia,

To go along with Daniel’s point, you might be better off with CTF parameters estimated by the Patch CTF routine: you can still account from defocus variations due to tilting, but get a more robust estimate by taking advantage of the bulk contrast rather than observations that arguably are not there at ~6A.



Hej Stefan,

thank you for your input. I have indeed put the defocus refine on ice for now. I did do patch CTF estimation at the beginning of my workflow and I increased the number of knots. Any other parameters to change that may help with the tilt?



Does the slope calculated through ctf-estimation match what you would expect from the tilt angle? If so, and to the best of my non-expert knowledge, it’s as good as it gets for this resolution.


Hmm, I’m not sure where I can read out the slope, do you mean this graph in patch CTF estimation?


Or do you mean the number for the tilt angle that comes up when you curate the exposures? The tilt angles CS gives are mostly below what we would expect (nominally, we tilted 20deg, I get numbers for most mics below that, more around 10-15deg), however, this all is complicated by the fact that we tilted a quantifoil grid which is not super stable.



Hi @ClaudiaKielkopf,

I think @stefan was referring to the tilt angle that you see in curate exposures. This tilt angle is estimated from the defocus landscape that was solved during CTF estimation, that indicates the varying height of particles in the ice across each micrograph.

Unfortunately I think this is the best that CTF estimation can get in this scenario.
If you look at the error landscapes that you posted first, you can see that the Y-axis of each landscape plot has a very small range. The values on that Y-axis are (more or less) in log-units of probability. This means that if the minimum of the landscape is only 1 unit below the maximum, there is only a difference in probability of exp(-1) = 0.36 which is not very significant. Ideally to faithfully detect the per particle defocus, there should be ~10 log units of difference in the landscape to be significant.