Puzzled - class percentages don't match number of particles used in refinement


So I have a dataset with 339000 particles. I perform an 8 class ab initio run, and one class looks good, with 26% (which would be 88000) of the particles according to the percentages given in the experiment tab. But when I refine this class, CryoSPARC only selects 11000 particles for refinement! Am I missing something here? And how do I chase down my runaway particles?

Is this due to the default class probability threshold?

If so, perhaps giving another percentage in parentheses, corresponding to the percentage of particles above the default class probability threshold, may be useful both when monitoring the classification and selecting classes for refinement?


Ahem, think I answered my own question - setting the class probability threshold to zero matches the percentages given on the experiments page.

Perhaps including the percentage of confidently assigned particles in each class on the experiment page or in the launch tab might be a useful statistic to monitor progress and tune classification parameters?


The default class probability threshold of 0.9 was problematic in my case, because it seemed I was systematically losing lower contrast, less-represented views - reducing the threshold seems to have fixed this issue.

Another idea/suggestion with regards to this - perhaps when editing the class probability threshold, cryoSPARC could give some feedback as to how many particles out of the total possible particles for that class this corresponds to?


Yes that is a bit confusing isn’t it, especially in cases like yours where the threshold matters a lot.
We’re starting design work right now of ‘CryoSPARC 2.0’ which will be more or less a rewrite of the whole package, so I’m putting this suggestion on our list there.