Hi,
the general strategy I follow for cleaning up a dataset is as follows:
-
Clean in 2D in relion (Hopefully soon in cryosparc??)
-
Import star into cryosparc; Perform ab initio heterogeneous reconstruction with two classes, and a class similarity parameter of 0.
-
This will usually give one junk class, and one nice, high res class (see examples attached).
-
Refine the good class, tuning the class similarity threshold to include more/less junk.
This works pretty well, but a two class ab initio run seems much slower than a one class run, and really what I want is a one class vs everything else/random noise run - but the class inclusion probability for a single class run is 1 (so tuning the class inclusion threshold during refinement does nothing).
I was wondering, would it be possible to include a kind of “null class” during a single-class run, so one could still tune on the probability that a given particle is actually in that class? Or some kind of correlation with the reference projection that one could use to tune inclusion of particles for refinement? Because at the moment, what I really want to know is - what is the probability that a given particle is in this (good) class, vs a random noise class, or some metric similar to that, rather than trying to find two well defined classes.
Apologies for the long winded post!
Cheers
Oli
