Precise meaning output Ab initio experiments


I’m puzzled a bit by the output of my ab initio reconstruction:

  • What does the color in the three axis-aligned slices tell me/what is the displayed quantity (and unit)?
  • What does the resolution output of an iteration tell me if I have a multi-class experiment? E.g., is it an average over all classes, the resolution of the best class or something else?
  • Why are the class percentages different on the experiments overview page than in the text-output of run n? The average % is averaged over all previous iterations I assume?



(somewhat related to Meaning of classification statistics?)

Hi Gideon,

  • The color in the three axis-aligned slices is just a colormap showing the “density” of the reconstructed structure. The colormap is on the right, and the units are essentially arbitrary density units (they depend on the electron dose, ice thickness, sample, detector, MTF, noise, etc etc) but they generally are not used as far as I know as real units, only relative units of density.
  • The resolution during ab-initio reconstruction is not a resolution estimate, but only stating the maximum resolution of information that has been used so far from the raw particles during the reconstruction.
  • The class percentages are different because during ab-initio reconstruction, we never see the whole dataset so the class percentages are only relative to the current mini-batch of data seen at each iteration. The percentages shown on the experiments overview page are the percentages relative to the total number of images seen by the algorithm so far.

Hope that helps!

Thanks Ali! Helps a lot