2D Classification Parameters Help

open

#1

Hi everyone,

I’m new to 2D classification and am getting weird results in my 2D classifications that I think is happening as a result of my parameters. For reference it is a set of about 400,000 particles. I have a picture for reference as well. Here are my parameters:

Window inner radius: 0
Window outer radius: 0.9
Number of 2D classes: 300
Max resolution: 8
Uncertrainty factor: 2
Circular mask diameter: 180
Recenter 2D classes: on
Recenter mask threshold: 0.2
Force Max over poses/shifts: on
Number of online-EM iterations: 30
Batchsize per class: 200
2D initial scale: 1
2D zeropad factor: 2
Ignore DC from image data: on
Use clamp-solvent to solve 2D classes: on
Use FRC based regularizer: on
Use full FRC: on

Does anyone have insight on how to improve parameters?


#2

Have tried keeping the parameters as default? Leave the mask diameter empty and see what the results look like?

Safi


#3

Safi,

Thanks for the reply. Yesterday I ran with more default settings and got slightly better results that look like the picture below. I was wondering if you had any tips on how to improve 2D classes that look like the ones below. They’re still slightly glossed over with features undefined.


#4

These are not the default parameters - you have “clamp solvent” on, which is rarely helpful. How does it look when you run with the default parameters?


#5

Here are results when i use all default


#6

I would try limiting the resolution used, and test switching the “Force/max” parameter off (so it marginalizes over poses and shifts), as well as increasing the number of iterations (e.g. to 40).

However I’m not sure that’s the issue - your classes look very weird. Maybe double check that all the preprocessing parameters are correct - e.g. Cs, kV, etc?

Cheers
Oli


#7

Hi @adtaheri, agreed with Oli’s recommendation to check your preprocessing parameters, especially the pixel size you inputted when importing the data. Thanks