Hello,
Could I ask for your thoughts on 3D classification without alignment?
I’m trying to classify a certain protein using 3D classification without alignment, where it’s known that two distinct states exist.
During the early iterations of F-EM, the data seems to be separated into two different classes, but as F-EM progresses, it eventually converges into a single class, and I’m no longer able to capture the two states.
I’ve tried various parameter settings such as filter resolution, number of classes, and hard classification, but none of them worked.
In this case, does it mean that only one state actually exists?
I would say that usually filter resolution is the best bet. Depending on the scale of the difference in your states it becomes harder to notice a change from class to class. So if this is not working, my suggestion would be to use a focus mask as an input, if you are not already. This way you can have the classes look at a specific spot where there may a change in the state.
Alternatively, as an off-the-path solution, you could use the 3D variability job (followed by 3D variability display) to create a movie of your protein. By importing the volumes to Chimera, you can see movement that way (when using the ‘simple’ output mode. Using the ‘cluster’ ouput mode will allow the particles to be separated into classes based on those states (usually with a histogram or PCA plot as representation). This is more or less a 3D classification job at that point.