Recently I have been playing around with the Hard Classification option in 3D classification. Since I see a small bump at ESS 2, I wanted to see if Hard Classification can help resolve the particles with low certainty.
However, after running the same 3D Classification with Hard Classification ON a few times, I noticed that the assigned particles vary greatly between runs. Is it because my uncertainty particles are not that much?
For example, I have 100k particles, and 3 runs of the same settings will give me: 60-40, 50-50, 90-10.
I think you should rather play with the parameter “class similarity”. And if your classes are very similar, keeping hard classification ON might actually be a bad idea, because you might be eliminating particles that could be useful in multiple reconstructions, especially in cases with continuous, subtle movements. Anyways, this is how I see it, but I’m curious to see what more experienced people will suggest.
If you haven’t tried 3DVA yet, I think it might give you useful insights. I use 3DFlex to see how a protein moves, then I make masks around the variable regions to be used in 3D classification, or masked 3DVA.
