Best way to test MotionCorrection and CTF estimation and compare results?


I am unaware if this has been discussed before, but I have a good data set that I want to get the most resolution out of. With so many different options for Motion Correction and CTF estimation, I was wondering what would be the most efficient way to test and compare the results between different methods? Patch Motion/Patch CTF estimation have been my standard but I am having “fear of missing out” knowing these other wrappers/methods are at my disposal.

I would like to just use a subset of my data (to reduce computational time and disk space) and run Motioncor2 and Patch Motion and then test these outputs with GCTF, CTFFIND4, and Patch CTF before running 2D classification.

I guess I would like to know if someone has any advice on how to execute this in the most efficient way possible. I have 5000 movies that comes with a final set of 1M particles. I was thinking of testing maybe 200 movies by importing all of the movies and then curating to just manually select 200 random movies. Maybe someone has an easier way to check the quality of the results that I am not thinking of.


Hi Karl,

I think the best approach might actually be to perform each method on the entire dataset, then extract your consensus set of particles and run a refinement - ultimately the resolution after refinement is the only metric that will tell you anything useful here - the internal metrics of each package are not directly comparable, and a 200 mic subset may not be big enough to allow you to tell if there is any difference in 3D - does that make sense?



I agree with Oli - 200 mics is probably not enough. 1,000 would probably give you a better idea, but then storage and time increases commensurately… using previously identified known-good particles to fix one variable is a good idea. However, the patch CTF estimation is very useful, particularly on lower magnification or tilt acquired data where there can be a lot of variation across the micrograph.

1 Like

Thanks Oli and rbs_sci,

I agree in that the internal metrics from each package are not directly reliable. I was hoping maybe there was a more empirical way instead of using the entire data set. For now, I will just stick with the method that has been working (Patch motion/Patch CTF estimation). I can determine if I need to push for higher resolution later on once I have the final map.