I have a basic/naive query about 2D classification in cryosparc:
Are the 2D classes simple averages of the particles that have their maximal probability in that class? Or are the classes weighted by the probability of assignment, in the same way as weighted backprojection is applied in 3D classification?
Currently the 2D classes are the weighted average of all particles by the probability of assignment, which includes particles that does not have their maximal probability in that class. In an upcoming release there will be an update to class2D that optionally does hard classification after the last iteration, which would make the classes simple averages of the particles that have maximal probability in the class. There will also be a new reconstruct2D job that reconstructs the classes using simple average.
This sounds great, thanks Serena!
Good to know!
This got me curious about how 2D class averages are produced in RELION: weighted reconstruction or max probability particles only? Does anybody know? @DanielAsarnow maybe?