Hello all,
I’m currently doing classification on small feature within 20 amino acids residues. It works pretty well and I got 4 different states---- 3 intermediate stats and 1 final state. Moreover, the final state gives us high resolution information which allows us to assign sidechains for each residue.
However, regardless of whether I apply force-hard classification or not, the particle distribution across these 4 states is roughly 25% each. The issue is the 4 states shouldn’t be evenly distributed under physiological condition or the majority of particles should be in final state in our sample condition.
Does the 4 states represent the averaging of different state or the real state?
1 Like
Did you adjust the class similarity parameter? It defaults to 0.5, try setting it to 0.001… also try increase number of particles per batch and number of iterations. For highly similar structures, 3D classification can take time to sort itself out (converge on the distinct/divergent classes, as it were)…
Thanks for your suggestion. Previously, I only tested the O-EM learning rate with initial values ranging from 0.1 to 0.9 in increments of 0.1, and I also tested particles per batch. I’ll now try additional settings based on your recommendation and keep you updated.
I also have a question regarding the parameter ‘Class similarity.’ Is the calculation of Class similarity based on the overall structure, or is it focused on the structure within the focus mask? As I used small focus mask, which only covers around 100 amino residues, for Classification.
Uncertain on that point, sorry.