Estimated error of translations/rotations

Hi!
I was wondering if there is a way to find out the estimated error of translations/rotations in Local and NU refinement procedures. A reviewer asked us for these.

Thanks,
Weida Liu

From our team:

In NU-refinement if the “Auto-marginalization” option is on (which it is by default) then there will be plots like this in the streamlog:

image

These are histograms where the x-axis is the standard deviation of the estimated posterior distribution over pose or shift, which is also referred to as the estimated error in alignment. The histograms are over particles in the dataset, showing the distribution of alignment errors.

This plot is not output in other jobs unfortunately and is not computed unless “Auto-marginalization” is on.

Hi wtempel,
Thank you very much for your explanation! Because a reviewer ask us for the value of estimated error of translations/rotations for every refinement result to show in the table. And I also do some Local refinement which need to be showed in the table. How can I get the precise value of these?
Thank you very much,
Weida Liu

Hi wtempel,
Thank you very much for your explanation! Because a reviewer ask us for the value of estimated error of translations/rotations for every refinement result to show in the table. And I also do some Local refinement which need to be showed in the table. How can I get the precise value of these?
Thank you very much,
Weida Liu

For results where these plots are available, the plots are the only representation of uncertainty in the poses output by cryoSPARC. The most readily available single numeric descriptor would be the mode of the distribution. If you chose to report the mode (instead of perhaps the entire plot), be sure to clearly state that the mode of the distribution is being reported. Keep in mind that a single number cannot convey all the information contained in the plots.
As we are unaware of these metrics being a “standard” validation measure, they may not be calculated by some jobs at all unless they are used by the algorithm.

1 Like

As an aside - such measures are also unavailable from Cistem/Frealign, which doesn’t marginalize. (The parameter file has a field called “sigma” but this is just the standard deviation of the particle image).

1 Like