Dark 2D Classes after Min over Scale


When I turn on ‘Min over scale after first iteration’ during 2D classfication, I get two different looking 2Dclasses, one light and one dark. Is the dark 2D class a ‘bad’ class? Should I keep the dark 2D classes especially if it seems to have high resolution features?

Screen Shot 2022-07-27 at 1.11.26 PM

Thank you,

Hi @joonpark,

I have never seen this so I’m not exactly sure what is going on here - my first guess would be that the scale refinement might be unstable, and for these particles it could be minimizing to a negative value (explaining the sign inversion).

Probably the easiest way to tell would be to select this class and have a look at the raw particles on the micrographs using inspect particle picks - do they look unusual in any way?


Hi @olibclarke,

Thank you for your comment. I selected the dark 2D classes and inspected the particles, but, overall, particles seem fine. It is avoiding ice contamination. There are very few wrong particle picks here and there, but I would assume that is the case for all 2D classes.

Is it safe to assume that these dark classes will not be negatively affect the 3D reconstruction?

Thank you,

I would say it is likely fine, so long as there is nothing else unusual about the micrographs where these particles are found.

1 Like

I have seen 2D classes like this from overfocused images.


Would you be able to post the images of all the 2d class averages

  • at the iteration where the dark class(es) appeared
  • after the final iteration
1 Like

In your experience, was it okay to include these classes in the 3D recontruction?

Dark classes seem to appear as early as iteration 7, but I could be wrong because most classes are still very fuzzy. It starts to appear clearly at iteration 9.

Iteration 9

Iteration 40

if they are actually overfocus then no, you don’t want them, but you should be able to tell that by looking at the micrographs (as contrast will be inverted)

If they don’t appear without minimizing the scale factors then my guess would be that it is an artefact of that. Why have you enabled this option, do you find that it makes a significant difference?

When I inspected the micrographs, the contrast was not inverted (dark particles on light background). Also, the data collection was done between -1 um and -2.5 um defocus, which was more or less consistent based on the estimated defocus value from CTF estimation. Based on these, I do not think they are overfocused.

Dark classes do not appear with the normal 2D classifications. However, when I compared results from with and without minimizing scale factors, it seems like I get better junk separation. Below is the 2D class results from without minimizing scale factors.

Here, I see clear and fuzzy 2D classes. Whereas with minimizing scale factors, I see some classes that are just fuzzy ball. Or is this fuzzy ball also an artifact??

I am facing a similar issue of light vs dark 2D classes upon merging particles from two datasets, have just posted herein seeking for advice please:

hoping to understand what I might doing wrong or how to fix