The function of 2D classification parameters


I have some questions about two parameters in 2D classification.

  1. Use clamp-solvent to solve 2D classes.
    What is a clamp-solvent? Is it a mask? How can it remove the streakings over the particle?

  2. Force Max over poses/shifts
    What do you mean by marginalize over poses? How can it help with small particles?


Hi @Chen,

  1. Solvent clamping is a new method for enforcing an empty background in 2D classes. It’s not the same as masking, and in cases where it’s useful we’ve found it to be better than masking. The idea is that during optimization, we enforce that the corners of each 2D class should be zero, that the 2D density should be non-negative, and the density should be smooth. Together this can reduce streaky artifacts in difficult cases. In average cases it doesn’t help too much (we think) so it’s off be default. NOTE: if you are going to use solvent clamping, you should also turn on class2D_nonneg to true.

  2. Marginalize here means that instead of picking a single orientation and shift for each particle vs. each class, we should take into account all orientations and shifts, weighted by their probability (so a more Bayesian method). It can help with small particles that have few features for alignment, by not allowing the optimizer to become overconfident in orientation assignment.

Hope this helps,