I am curious if there are situations where it makes sense to change the defaults regarding re-centering 2D classes during 2D classification:
Centering ring-like or cage-like particles:
I work a lot with nanoparticles de novo cages, where the highest density in 2d class averages is in a ring, and the center is often very low density. I am not sure if the term intensity is synonymous with density in this context. It seems sometimes the particles are being centered at points around the ring with the highest density, rather than at the true center of the nanoparticles.
Weeding out ice from the data set:
While the 2D classification runs, before the final iteration, I often see classes containing spheres of ice that clearly have high density at the very center. Unfortunately, these do not get sorted out. After being sorted into separate classes intially, they are folded back into the good classes. I am curious if this centering by intensity is causing the ice particles to stay in the data set?
These are the parameters I am referencing:
“Recenter 2D classes (ON): Whether or not to re-center 2D class references at every iteration to avoid drift of density away from the center of the box. This option is often important to keep classes centered and avoid artefacts near the edges of the box.
Re-center mask threshold (0.2): 2D classes are recentered by computing the center-of-mass (COM) of pixels that are above this threshold value. The threshold is relative to the maximum density value in the reference, so 0.2 means pixels with greater than 20% of the maximum density.”
Thanks so much for taking the time to read!