results of ctf-estimation differ greatly between ctffind3, xmipp, robem and ctffind3 from command line

Submitted by dieter on Fri, 2010-09-17 17:56

Forums

ctffind & ctftilt

Hi,
as in my first post in the signature user group I first want to thank the developers for making available this fantastic software!
We are just trying out xmipp and have compared different ways of calculating the ctf. The results differ a lot when using ctffind3 (latest version) from command line, from within xmipp, when using the xmipp built-in function, and with robem. I find it quite striking that ctffind3 from command line gives a large number of rings extending far beyond the credible range (one does not see anything like that in the power spectrum). On the other hand, xmipp gives far less and even less then the rings distinguishable in the power spectrum. The numerical values also differ by some 5 to 10%. Can anybody advise me how to proceed? Does it make sense to first use ctffind3 to just throw out the bad images and then to do a careful visual fitting with robem? Are these differences insignificant?

Thanks a lot, Dieter

niko Sat, 2010-09-18 08:16

I am not sure I understand what you mean when you say that you see more rings when using ctffind3. Are you talking about the calculated power spectrum (right side of the montage that ctffind3 writes out)? The number of rings you can see might depend on the way the power spectrum is calculated. In ctffind3, the image is divided into many smaller areas (for example, 128 x 128 pixels) and the power spectra of all areas are then averaged. To see rings at higher resolution, it might also be important to subtract the background. If the background is not subtracted, you might have to play with the gray levels of your display program to see the weaker intensities at higher resolution. I do not know how the power spectra are calculated in xmipp or robem. However, the defocus values found by ctffind3 and ctftilt should be good enough to get you to better than 4 Angstrom resolution, based on the near-atomic resolution structures published with ctffind3/ctftilt. If you are trying to get to 4 Angstrom resolution, differences of 5 - 10% will be significant.