Improving Tomograms with Ribosomes as Ficuials

[Euan William Pyle]

Hi WARP Friends,

For a while, I've been refining ribosomes in M. I'm not actually interested in the structure: I'm using them as 'fiducials' to improve the alignments of the tomograms they're in.

I'm starting to second guess myself, and feel that I've over-refined whilst doing this, in practice making some of the tomograms worse.

My question is: when refining particles in M, with the aim of improving tomogram quality to pick/refine other particles, how far do people go?

I've heard suggestions to only use image_warp 1x1 (2x2 maxium), refine_particles and repeat that a few times. Do people also have success adding: refine_stageangles, refine_volumewarp? Or does this run the risk of optimising the tomogram alignments only around the particles of interest to the detriment of regions elsewhere?

Apologies if this has been asked before!

Thanks

Euan

[Alister Burt]

Hi Euan,

It’s definitely possible to overfit deformation models to a particular species at the expense of the rest of the tomogram. This doesn’t mean the reconstruction of your species is overfit, it can mean it’s correct where your particles are and incorrect elsewhere. All models are wrong, some are useful!

I’ve seen this before in tomograms with microtubules at the center, after refinement the microtubules looked amazing but the edges of the field of view got worse.

Repeating 1x1 or 2x2 image warp + particle poses is a good general strategy, whether you can go further will depend on

- data quality

- number of particles

- spatial distribution of those particles

If you have multiple particle sets within the same volume you can assess this quantitatively by

- refining deformation models on particle set 1

- refining only particle poses on set 2

- assessing resolution of set 2

You could approximate the same thing by comparing template matching peak heights pre/post refinement for particle set 2.

If you don’t have multiple species you could split your ribosome species into two artificially, Dimitry did an experiment like this in the M paper although I think it was only for SPA 

Some notes:

- There is no such thing as image warping in the microscope, you have a 3D sample which is deforming in 3D. 

- Most of the time image warping can approximate this deformation. (All models are wrong, some are useful)

- Image warping can’t account for 3D deformations where particles at different z heights move in different directions. 

- Volume warping can account for these deformations but requires images from different projection directions to fit so necessarily has temporal resolution. 

- image warping is fit independently per image so the results are “local” in time. 

- If you’re going to fit volume warping you should have particles well distributed in all 3 spatial dimensions of your volume. 

- Whether you get any benefit depends on how well image warping was able to approximate the actual 3D deformation of the sample and your particle distribution

Stage angles are usually pretty good, I don’t think your tomogram would benefit much from refining them

Cheers,

Alister

Sent from mobile - apologies for brevity

On Mar 5, 2025, at 04:47, Euan William Pyle <euan...@embl.de> wrote:

Hi WARP Friends,

--
You received this message because you are subscribed to the Google Groups "Warp" group.
To unsubscribe from this group and stop receiving emails from it, send an email to warp-em+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/warp-em/42520ee8-bbb0-4c94-855a-1b1e79b8ad80n%40googlegroups.com.

[Euan William Pyle]

Hi Alister,

Thanks for the super detailed reply! My dataset is quite huge (and varied in terms of thickness, and number of ribosomes in the sample, and their spatial distribution), so based on what you've said, I think I will stick to simple image warp / refine particle poses to improve the tomograms. I think with volume warping, there are too many tomograms that don't have good spatial distribution in all 3D. I like this idea of splitting the particle sets to assess the improvement of the tomos, I should have spotted that in the WARP/M papers, thanks!

Thanks

Euan

[Euan William Pyle]

A follow-up so the historians of the future can see the result of this thread ;) 

I went back to the original (pre M) tomogram state and ran M again. This time I ONLY used image_warp 1x1 and refine_particles. I ran this over and over until the resolution improvements became what I considered too small to care about (0.2A). Finally, as I'd reached roughly 5.5A, I considered this good enough for defocus refinement, so I did that, once only. 

The result: I reached the same resolution as I did before (I am averaging ribosomes here so it is slightly easier to reach high res), but my tomograms have definitely improved. I'm judging quality by: how well defined microtubule subunits are, how well defined membrane leaflets are, etc. etc. 

This obviously isn't a quanitative study, so take with a grain of salt, but in future when using ribosomes as fiducials, I'll almost certainly be super conservative with M parameters (unless someone else has had different experiences who wants to chime in!)

Euan

[Oleksiy Kovtun]

Hi Euan and Alister,

Thank you for a discussion on this topic and sharing your observations. Very interesting. 

Best.

Oleksiy 

To view this discussion visit https://groups.google.com/d/msgid/warp-em/15e1a928-d92d-461c-a57b-271e4b7765e1n%40googlegroups.com.

[Yunjie Chang]

Hi Euan and Alister,

This topic is very helpful. I have also thought about using ribosomes as fiducials before, but have not really tried it.

Maybe I missed some functions of warp or M.

I want to ask how to generate new tomograms after running the refinement in M? Is there a function in warp/M to use the refined particle positions to reconstruct tomograms? Then will this also generate a new xf file for each tilt series?

Or should I select some good ribosome particles based on the refinement results, and feed them to etomo as fiducials and let etomo align the tilt series again?

Thanks

Yunjie

[Euan Pyle]

Hi Yunjie,

Thankfully, it’s quite simple to do. Once you have run M, and have a nice ribosome structure, just re-run the Tomogram Reconstruction step that you did earlier. I believe that M updates the .xml files in warp_tiltseries with the improved alignments if I’m not mistaken. I usually save my ‚old‘ tomograms in a new directory for comparison purposes before running the Tomogram reconstruction step.

Best

Euan 

On 29 Mar 2025, at 09:17, Yunjie Chang <chang...@gmail.com> wrote:

Hi Euan and Alister,

To view this discussion visit https://groups.google.com/d/msgid/warp-em/6df5b32c-ae94-4668-a431-b409ae4ab845n%40googlegroups.com.

[Alister Burt]

Hi both,

Euan is right - few thoughts below

Warp/M is a “state machine”, when you run any program the xml files are updated and any relevant output files are produced. The xml files for tilt series contain parameters which describe reconstruction geometry, those are updated when you run M.

An xf file is contains IMOD’s description of reconstruction geometry which is not as flexible as M’s so producing a “new xf file”doesn’t make sense

Cheers,

Alister

Sent from mobile - apologies for brevity

On Mar 29, 2025, at 06:02, Euan Pyle <euan...@embl.de> wrote:



To view this discussion visit https://groups.google.com/d/msgid/warp-em/1E828697-16A4-4342-AAD4-975CEBEBA481%40embl.de.

[Yunjie Chang]

Thanks Euan and Alister. 

This makes sense to me, will try on my own data.

Best

Yunjie Chang

Alister Burt <alist...@gmail.com>于2025年3月29日 周六22:49写道：

[Peter Van]

Hi all,
The M can update the .xml files in warp_tiltseries with improved alignments. I have a couple of follow‑up questions:
Can Warp/M export the updated, aligned tilt series so that we can perform alternative tomogram reconstruction methods (e.g., SIRT or SART) in other software packages?
If yes, what is the recommended strategy or command to do this?

Best,

Peter