Noise model plot meaning

Ho. I am new to CS (and tò cryoem in general). I do not understand the noise model plot display ed inab inizio reconstrution.what’s the meaning of sigma and error plots? Is theresamo kind of explanation somewhere?

Hi @stefanocapaldi,

Welcome to the forum! The noise model (sigma) plot shows us an estimate of the non-structure noise present in the images, as a function of spatial frequency. It is estimated via computing the “residual” in each image – this is the squared difference between the raw data, and the CTF-corrupted projection of the density. If you are familiar with Scheres, 2012, it is very similar to equation (4) for estimation of sigma.

The sigma we compute & plot is this squared difference, averaged over all images, and also averaged over all frequencies at a given resolution shell (we call this a “symmetric” or “circularly symmetric” noise model). The plot for error shows this quantity before circularly-averaging, but the differences between sigma and error are usually insignificant.



Hi @mmclean ,
got it. Thank you for the explanation.


can I ask how I might use this plot in 2D or 3D to guide my data processing? For example there is an option in 2D to use the noise model, but I’m not sure when I might use this.



Hi @orangeboomerang,

In general, the noise model plots are most useful for diagnostic purposes to developers rather than users, and we don’t have any strong examples of how it may inform data processing. There is one exception I can think of, which was discussed in this thread where prevention of the annealing of the noise model (i.e. using a white noise model with inflated variance) improved 2D classification results.