3D classification gives similar percentage of particles of each class

[Wendy]

Hi，
I tried to use the 3D classification to classify different conformations. It indeed gave me different conformations. However, it seems every class have a similar percentage of particles. I tried simple, input, PCA mode, and all gave me a similar result. I also tried to 3D classifiaction in relion, the percentage of different class is totally different. Does any one have good suggestions to change the 3D classification parameters in cryosparc?

simple mode

PCA mode

input mode

Thanks,
Wendy

[cbeck]

Hi Wendy!

In my experience, just turning on “Force Hard Classification” really helps with preventing particles from being equally distributed across all classes. You can find some further information on what parameters to play with in this thread:

3D Classification strategy: Classify until exhaustive?

Hi everyone! Today I would like to ask a question on your strategy involving the use of 3D Classification for peeling may conformations from a dataset of a wildly active protein. I have been using 3D Classification on 800k particles (selected by many rounds of 2D classification to remove junk particles) and split into 4 classes consisting of 180k, 250k, 198k, 203k particles each. Then, I realize that each class can be split into 2-3 more classes. For one of those subclasses consisting of 40k p…

Cheers,
cbeck

[Wendy]

Hi cbeck,

Actually, I have tried the parameter “Force Hard Classification”. It not works for me. still got the same particle distribution.

with force hard classification.

without force hard classification

Thanks,
Wendy

[Mark-A-Nakasone]

Hi @Wendy for 3D Class, what is your input for Filter resolution (Å)?

If O-EM learning rate init is low 0-0.4 less diverse classes, if it is high 0.75-1.0 then you will get very diverse classes.

you could try different filtering depending on you resolution and see if that does anything. sometimes this is what happens. keep force hard classification on.

[Wendy]

Hi Mark,

My Filter resolution is 6 Å, I have also tested 4Å. it didn’t make big difference.

For O-EM learning rate init, i used the default parameter. I will try to increase it to see what will happpen.

Thanks,
Wendy

[Wendy]

Hi Mark,

I tested different parameters, and each configuration produced different results, with significantly varying ratios across the different classes. Which result should I trust?

filter resolution 4, O-EM learning ratio=0.4

filter resolution 4, O-EM learning ratio=0.6

filter resolution 4, O-EM learning ratio=1

Thanks,
Wendy

[Mark-A-Nakasone]

Hi @Wendy Great to see the changes from your first post on 3D Classification.

Can you simply take the particles from the 3D class and run Homogeneous reconstruction only. You may be able to see a somewhat high resolution map provided the particles going in were the full box size, if not just take the particle stack from 3D classification, re-extract at full box size, and then homogeneous reconstruct.

It can be hard to say which is best from the 3D classification job alone given the filter.

Also very important to see if these per-class reconstructions are radically different.