Hi,
The class probability threshold was very useful to fine-tune selection of particle sets after heterogeneous refinement in cryosparc v0 - it is a shame to see it gone in v2. Is there any possibility of bringing it back?
Cheers
Oli
Hi @olibclarke, thanks for your feedback, we are on this and it will be part of a future deployment.
1 Like
Hi @spunjani - any chance this is coming back? I still miss having this available in v2 - it was very helpful in v0 to tweak selection of classes in order to find the best subset of particles, and just allows a bit more control over particle selection
Oli
Perhaps it could be incorporated into the particle subsets tool?
For anyone (or future me) searching this - for now, the way to do this is using the --minphic flag in csparc2star.py from the pyem package. E.g., to only keep particles with posterior prob >0.99 from class3 of a particles.cs file from heterogeneous refinement:
csparc2star.py cryosparc_P1_J556_final_particles.cs cryosparc_P1_J556_final_particles.star --minphic 0.99 --class 3
Cheers
Oli
2 Likes
I see this is back now in 2.15, thank you!!
1 Like
@olibclarke sorry to ask: could you maybe say something on how you use the threshold and what decisions help you with ? Many thanks !!
Hi @marino-j - sometimes using a more rigorous threshold than the default gives a cleaner set of particles with less junk. This can also be helpful in generating a particle set for training a machine learning picker such as topaz.
It would be useful to be also be able to use a lower threshold than the default, in cases where the probability distribution is broad and the default threshold may be a bit aggressive - this was possible in csparc v1, where the probability threshold was applied at the same time as class selection, but not in v2, where it is applied in a second step as a separate job type.
Cheers
Oli.