ctffind4 output power spectrum

Submitted by Grigory on Mon, 2016-02-15 09:23

Forums

ctffind & ctftilt

Dear Alexis,

I have done CTF estimation with CTFFIND4 and was looking through the results. While the fit looks OK, I was wondering whether is it normal to have such strange "BG circle" on the .ctf file (see the picture attached) after CTF estimation? I have attached original amplitude PS of the same micrograph for comparison. Maybe background subtraction went wrong in ctffind or this is OK? I did put a question mark on the image where it looks like some sort of a mask was applied.. though I used the full range from 30A to Nyquist for CTF estimation (as can be seen on the theoretical spectrum).

Alexis Mon, 2016-02-15 09:46

Dear Grigory,

Great question.

Ctffind4 tries to boost the contrast of experimental Thon rings in your amplitude spectrum. To do this, it "guesses" the expected positions of peaks and troughs in the Thon ring pattern, based on the results of the fit, and then renormaises grey values in those rings/ellipses. This often works quite well, except in cases where there is significant aliasing of the CTF function (i.e. CTF oscillations get close to or smaller than 2-4 pixels long).

In your case, there is significant aliasing of the CTF from about 1/4 of Nyquist onwards. This means that the "contrast" boosting algorithm will not work properly, hence the artefacts you are pointing too.

More importantly, it means the results of your fit a probably sub optimal. You should consider resampling the input data to a larger pixel size ("binning"), so that your Thon rings extend closer to the Nyquist during fitting. And/Or use a larger box size for the fitting, so that aliasing doesn't happen so quickly.

At the very least, I think you should consider only doing the fitting to, in your case, something like 1/4 Nyquist. I don't think fitting against aliased CTF functions will give good results.

The next version of Ctffind (not ready for release yet) has a feature which detects aliasing and gives a warning to the user if aliasing is detected within the fitting range.

HTH
Alexis