Ewald Sphere Correction

Dear colleagues,

I try to follow the manual to test the Ewald sphere correction protocol.

My first Non-uniform refinement reconstruction ended up with 2.07 A for masked case

Then I run the Homogeneous only reconstruction using the particles of this job.

But I got much worse results - 3.48 / 3.43 for unmasked and masked cases correspondingly.

So why did that happen? Am I doing something wrong?
How to correct that?

Thank you

Sincerely,
Dmitry

Maybe a stupid question, but did you make sure you had the same symmetry enforced for both?

1 Like

hello @olibclarke.

  • yes “O” symmetry . I checked this by the log + run the C1 and it finished even at worth condition 4.06. Maybe I need to rerun the 3D non-uniform refinement since the previous one was done with an older version of cryosparc…

Hmmm in that case that is very strange, I’m out of ideas, maybe a bug of some kind

1 Like

yeah…seems that I am good in finding bugs )). I am will test it on the other dateset

Hi @Dmitry,

This is interesting, and I struggled to find datasets that benefitted from both non-uniform refinement and Ewald sphere correction simultaneously, so combining the two features wasn’t very high priority during development. I’m glad to know that such datasets exist, and that we should definitely consider combining these two features!

Likely the reason you are finding higher resolution with non-uniform refinement is because marginalization is used by default in the non-uniform refinement job. So, orientations and shifts are integrated over when updating the structure each iteration, because we can estimate the probability distribution over poses as part of the refinement protocol. This is in contrast to the reconstruction job; we only save a single point estimate of the final orientation (it is impractical to write out a distribution over poses). The distinction is highlighted in Scheres’s 2012 paper on the bayesian view of cryo-em structure determination. All in all, this means that it’s unfortunately not an apples-to-apples comparison to compare the output volume of non-uniform refinement to the output of homogeneous reconstruction only.

If you are able to, could you take the particles from the non-uniform refinement job and then:

  • Run homogeneous reconstruction only without Ewald sphere correction enabled
  • Run homogeneous reconstruction only with Ewald sphere correction enabled (with the better curvature sign that you found)

and let me know which of the two produces a better resolution, and by how much?

Best,
Michael

1 Like

Dear @mmclean,

Thank you for your answer.

I was able to reproduce the Ewald sphere correction from the tutorial. But for my current dataset, the Ewald sphere correction did not improve the map. However, I defined that my map should have a negative Ewald sphere sign. The resolution stays almost the same (even a bit worse than the one resulting from 3D non-uniform alignment: 1,91A Ewalp sphere corrected vs 1.86A non-uniform refinement). So what could you suggest to do for the map improvement or I reached the possible end, and there is no more information left to reconstruct?

Kind regards,
Dmitry

Hi @Dmitry,

How much was the improvement when you compared negative curvature to no curvature reconstructions? Based on your workflow, it does seem like there isn’t much else to do in the way of Ewald sphere correction to help improve resolution, until it can be incorporated into non-uniform refinement.

You might also find it beneficial to compare the 3D maps as the tutorial does (sharpened to same b-factors), to check if the added few decimals of resolution corresponds to a noticeable map quality increase.

Best,
Michael

1 Like

Hello @mmclean,

So the starting point - non-uniform refinement

No Ewald sphere curvature sign:

Negative Ewald sphere curvature:

Did the NU-refinement map end up looking like the better of the maps (vs homogenous reconstruction-only and HR with negative EWS)? Just curious because I happened upon this thread doing the same workflow. I’m working with a large molecule that comes out to 3.59A overall after NU-refinement so technically EWS shouldn’t be useful, but it’s flexible and has an area of much higher resolution on one side so I thought I would try anyway.

Homogenous reconstruction-only after NU-refinement resulted in a slightly worse map (difference of 0.22A) and then there was no visible map difference between that and EWS correction in both neg/pos instances, so in my case it didn’t help after all.

1 Like

yes, the NU-refinement map in most of my cases looked a bit better