Ctffind4 version 4.1.13 worse than 4.1.4? Bug?

Submitted by mceachenj on Fri, 2020-01-03 09:39

Forums

ctffind & ctftilt

Hi all,

I updated my ctffind4 installation from veriosn 4.1.4 to version 4.1.13. Using the same parameters as before the update, I now get worse defocus estimations.

Details: I am estimating the defocus of a tomographic tilt series. Each tilt image was drift corrected but not dose-weighted. The target defocus was 3.5um. Around small tilt-angles the estimation seems ok. At higher tilt angles the 4.1.13 version is completely off while the 4.1.4 version gives good results. I am using the same parameters that gave good results on many other tilt series. You can find details about the chosen parameters and a plot of the defocus in the attached image (the tilt numbers 1-41 in the plots represent the tilts from -60 degrees to +60).

Am I making a mistake in the parameter selection or is this maybe a bug?

Thanks a lot for your help,
Stefano

himesb Wed, 2019-02-20 11:59

Hi Stefano,

Perhaps, unrelated to your bug, you may find more success in general with a tilt-series if you:

1) Limit your high-resolution fitting, perhaps to ~8-10 Ang

2) Restrict your search range a bit more judiciously. This will help you to remain in a closer ball-park in the initial fitting. (Which also could be related to the differences you are seeing here.) Maybe as low as 25000 and as high as 50000

3) Assuming decent information (particles) in the center, it can be helpful to first trim your input stack around the tilt-axis to exclude regions that are going to degrade the average at higher tilts. Using just the 1024 pixels around the tilt-axis for example.

HTH

Ben

Alexis Wed, 2019-02-20 10:53

Hi Stefano,

It could be that 4.1.13 has a bug which wasn't in 4.1.4. It's also possible there was a bug in 4.1.4 with a useful side effect (in your case) that was removed in 4.1.13. Hard to tell just from this.

The only slightly unusual setting in your parameters is the low resolution limit. You have 50 Å, whereas I almost never deviate from 30 Å for this parameter during testing and development. Could you see if the problem reproduces with 30 Å? If so, please share with me two example micrographs: one that works well and one that only works in 4.1.4. You could do this via this forum (by editing your original post in this thread) or over email directly with me.

Thanks for carefully testing ctffind!

Cheers,
Alexis

stefano Wed, 2019-02-20 12:38

Hi Alexis,

Thank you for the very quick reply!

When using 30 Å as a lower resolution limit, the results look similar, if not even worse. With tomographic data I always had better results using 50 Å instead of the default 30 Å for some reason.

I tested both ctffind4 versions now also on Florian Schur's flawless HIV dataset (EMPIAR-10164) in order to exclude data specific problems. The 4.1.13 failed less times than on my own dataset but I could still see that at high tilts there are often problems which I didn't have with the older 4.1.4 version. Note that this issue appears at much lower titls in my own dataset.

I sent you two tilts from EMPIAR-10164 by e-mail to illustrate the problem (they were too large to upload here). I analyzed them using ctffind's default inputs and a pixelsize of 1.35 Å. Target defocus is 4um. If anybody else wants to give it a try, the tilts are:

- tilt35.mrc: Tilt nr. 35 (+45 degrees) of stack 1 with correct defocus estimate in both versions (~4um)
- tilt36.mrc: Tilt nr. 36 (+48 degrees) of stack 1 with correct defocus in 4.1.4 (~4um) and wrong defocus in 4.1.13 (~1.5um)

Thanks a lot for looking into it.

Best,
Stefano

Alexis Wed, 2019-02-20 12:43

Hi Stefano,

Thanks for all this. I will investigate this, but to be honest it may take me a while to get to it. I tend to work on ctffind in spurts, and my last bout of activity was late December / early January. So it will probably be a few weeks before I get to this. Sorry if this quite a long time.

I'll look out for your email (not received it yet) and keep it in my inbox.

Cheers,
Alexis