Dust removal during particle extraction?

[olibclarke]

Hi,

Consider a situation like this (denoised & not denoised):

The little black dots are (I think) small gold particles shed from the grid. Where they end up in an extracted box, these can perturb alignment, and lead to 2D classes like this:

Relion has an option to deal with this kind of situation, by which pixels above and below a user-configurable number of std. devs from the mean are replaced with the mean value (or noise distributed around the mean value - not sure).

I understand from previous posts on the forum that particle extraction in CS does something similar, but it is not user-configurable. Would it be possible to make it user adjustable, in order to allow rescuing more usable data in situations like this? Would also help with proteins bound to gold nanoparticles, metal clusters etc.

Cheers
Oli

[vincent]

Hi Oli,

thanks for reporting, it’s interesting. It’s quite a lot of gold, do you see this often?
Does it impact 3D reconstruction if you go directly to ab-initio?

Thanks
Vincent

[olibclarke]

Not often but enough that I would like a better way to deal with such data! Yes, high contrast junk like this affects ab initio badly.

[apunjani]

Hi @olibclarke thanks for the report, definitely the presence of strong contrast “dots” would affect abinitio and pose alignment generally.
In CryoSPARC, we do hot pixel rejection at the movie frame stage, but not within micrographs after motion correction. Since these dots are not detector defects, they probably wouldn’t get picked up by the frame-level hot pixel detection (because these pixels would be dark, i.e. lower electron counts, not hot).

We’ve recorded this topic internally for investigation.
Have you tried removing the dots in Relion using the stdev pixel replacement you mentioned? It would be good to know if you found that to be a workable solution.
Thanks!

[olibclarke]

Hi Ali - I haven’t tried re-extracting in relion in this instance to compare, but it is a good point - I will give it a go and do a direct comparison by Class2D!

Cheers
Oli

[August]

I could also use this feature even if it is implemented in the micrograph junk removal job. There is a common issue with some Falcon 4 cameras where there are man bright spots, there are too many spots of varying sizes and intensities to use a defects file. The dust removal feature would be very useful.

[rbs_sci]

The sensor passed qualification with so many defects that it is causing such severe issues? I’ve spoken with a number of facilities in Japan with Falcon 4s and none have mentioned such issues…

[August]

it is something similar to this EER defect & gain reference problem

typically ok, but the bright spots become a problem with denoising and the ripples cause problems with small particles and ab initio.

[rbs_sci]

Hm. Curious. I’ve seen that ringing effect once or twice… but not on our Falcons. If still under warranty or service contract, I’d suggest requesting that the sensor is checked and replaced if necessary - also double check the defect rate allowance; we have two Falcon 4s, and one has a really low defect count while the other is/was apparently close to failing qualifying due to defect count (if comments from the engineer at the time are correct)… and I’ve not seen it so bad to impact picking that severely, even on that camera.

Interesting. I’ll keep a closer eye out.

[olibclarke]

We also see this on data collected on a falcon 4 (collaborator collected the data at a facility I am not familiar with - Krios with Falcon 4)

EDIT:

Just remembered in previous cases where we had this issue, running cisTEM remove_outlier_pixels on the gain reference fixed the issue - maybe worth a try here too?