Warp reconstruct tomograms creates a lot of empty space

378 views
Skip to first unread message

Felix Lange

unread,
Jan 29, 2024, 2:27:11 PM1/29/24
to Warp
Hi guys,

I am using dynamo_autoalign (thanks Alister!) to process tilt series for Warp. But during reconstruction, I have to use fairly large dimensions to capture the whole information of the tomograms. For example 4500 x 6500 x 1500 pixels (images from K3 with tilt axis of 95.4° - enclosed you can find an example image) Especially in z, the tomograms could fit into a much smaller volume but whenever I reduce the z-height, the information in the tomogram will be truncated from the bottom and not equally from top and bottom. Is there a way to handle the "centering" a little better/smoother?
Thank you for your help and all the best,
Felix
Capture.PNG 


Alister Burt

unread,
Jan 29, 2024, 5:51:09 PM1/29/24
to Felix Lange, Warp
Hey Felix,

Tomogram looks cool, glad the scripts are useful for you!

What benefit do you hope to get from centering? If only visual, I would probably just go ahead and crop the volume for visualisation.

It is definitely possible to encode an additional shift of your volume in the XF file, accounting for the extra 3D shift in your sequence of rotations/translations that make up the projection model. 
This will be a bit finicky and require a pretty deep understanding of how to interpret/manipulate the metadata coming from IMOD.

Good luck if you end up giving it a go!

Alister



--
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 on the web visit https://groups.google.com/d/msgid/warp-em/f5647adb-2242-4a1d-8dfb-68fb1918cc0en%40googlegroups.com.

Felix Lange

unread,
Jan 30, 2024, 3:06:02 AM1/30/24
to Warp
Hi Alister,

Thank you for this suggestions. My question was purely motivated to first limit the storage requirements for each individual tomogram but also to speed up potential template matching routines (Dynamo) which will spend quite some time on just processing empty space. 
As far as I am concerned, any cropping would need to be reflected in particle shifts for later subtomogram extraction in Warp - for pure visualization I agree with you! 

I noticed when you set the z-height fairly small the "truncated" information at low z-values will actually be added in the top of the tomogram. For example you will have halve of a liposome at z-height 0-20 and the "rest" of the liposome ends up on lets say z-height of140-160. But I have not explored these effects in much detail.

For your suggested approach one has to look into the XF file in IMOD and see how it changes after tomogram positioning (only z-shift and no x-tilt) - I will give this a go in the future.
All the best,

Felix

Lorenzo Gaifas

unread,
Jan 30, 2024, 5:51:45 AM1/30/24
to Felix Lange, Warp
Hi all,

It looks to me like you have some wiggle room there, and you could make the thickness almost 2/3 of what you have now without cutting into your data:

Untitled.png
Otherwise, ss far as I know, there is no way to get the "tomogram positioning" info from IMOD to be used by warp (see http://warpem.com/warp/?page_id=378, section "Create TomoSTAR files from IMOD’s results").

Until now, I personally just bit the bullet and had bigger tomograms than I wanted, because dealing with the coordinate shifts is very error prone and a bit of a pain. I'd love to hear if you figure out an ergonomic solution :)

Best,
Lorenzo


Higor Rosa

unread,
Jan 30, 2024, 6:32:59 PM1/30/24
to Warp
Hello everyone,

I’d like to share a tip that could be beneficial to others, particularly in relation to tomogram positioning. A while back, Digvijay shared a method on this forum for finding and updating tilt angles (https://groups.google.com/g/warp-em/c/hUxPsXR07t0/m/1dvWhpFZBwAJ). This is one of the methods to do it and it is based on the DoseRate, which can be used to determine the 0 tilt due to smaller thickness and higher dose rate at this specific angle. Essentially, it adjusts the tilt angles in the mdoc file you supply to Warp before creating the .st files. This adjustment helps to minimize the large tilt offset between the stage angle and the actual lamella angle, which can occur during lamella preparation, for example.

Although this adjustment alone doesn’t drastically improve results, it does help to slightly reduce the extended thickness required for tomograms in Z. This is somewhat similar to tomogram positioning, but as it occurs at a much earlier stage in the process it's actually "accepted" by Warp. I’ve experimented with processing tomograms reconstructed using this method, and so far, it hasn’t introduced any artifacts that I was initially concerned about. I’ve been able to perform template matching/DL segmentation and all the M-Warp-Relion pipeline STA steps without any issues. I could even refine these tilt angles in M without any noticeable drawbacks.

Interestingly, I’ve noticed that my AreTomo alignments also seem to improve when using these “tilt-angle corrected mdocs”. My theory is that having a good initial “reference” where the 0 angle is at the pre-tilt angle (at thinner thickness and higher contrast) benefits the iterative alignment. This “higher contrast” at this tilt seems to enhance all subsequent alignments (as they rely on the first/previous one). I’ve been using this method even for tomograms I’ve reconstructed outside of Warp. If you’re not using Warp, this approach is somewhat akin to activating the AreTomo TiltCor flag. However, at least from what I've seen in the AreTomo log, it seems to improve the alignments even before the tilt-offset refinement and correction is started, maybe benefitting the whole process.

Still this only effectively affects the "Z-axis" rotation of the tomogram. For the x-axis, I think this comes mainly due to a rotation between the microscope tilt-axis and the actual "lamella-axis" (for those using PACEtomo, this would be the lamella "rotation"). This can also be further affected if your tilt axis is not properly calibrated, but for this latter, you can actually correct this using the flag TiltAxis in AreTomo. However, for the actual difference due to the "lamella" rotation, one would have to dig deeper into the .XF files to compensate afterward. In practice, you can avoid this by loading your grids in the "correct orientation", and in the future, we hope this will be overcome by grids/autoloaders that preserve the grid orientation from the FIB-SEM (Arctis will have with similar AutoLoader/CompuStage from Krios!?). Anyway... I'm actually really interested to see if someone has actually tried to correct this x-axis rotation a posteriori and has seen any actual benefit from it that would be worth a try.

Hope it helps!

All the Best,

Higor

-

Ph.D. Student @Mattei Research Team (EMBL - Imaging Centre)

Felix Lange

unread,
Feb 5, 2024, 3:46:56 AM2/5/24
to Warp
Hi guys,

thank you all for the suggestions! 
@Higor: Did you have a look on this tool from the Grotjahn lab:

https://github.com/GrotjahnLab/warp_lamella_adapters

Maybe this accomplishes pretty much the same? 

Cheers,

Felix

Reply all
Reply to author
Forward
0 new messages