Reconstructed tomograms have low contrast (Aretomo1/2 aligned)

[Kyle Messina]

After struggling a bit I'm finally getting the hang of WarpTools and its making more sense overall.  Right now I'm running into an issue with WarpTools reconstruction.  I'm able to get well aligned tomograms but they have very low contrast and I'm not sure why.  

I'm using Aretomo1/2 to do the alignments (since the wrappers did not work well) and using the ts_import_alignments command to import the alignments.  The Aretomo1/2 have good alignments and good contrast but the WarpTools reconstruction look like the included images (hopefully they show up, first time posting images to a Google group) but with good alignments.

At first I thought Aretomo version compatibility was an issue so I tried different Aretomo versions for alignments but the only thing that changed was how good the alignments were. I did see a post about a similar issue but their fix (making sure their .xf have alignment data instead of just 0/1s) doesn't look like it applies to me.  So I'm not sure where to go, (or if the low contrast matters)! 

If it matters, I'm using the EMPIAR 11830 data set which consists of EER files. I'm not used to using EER files for tomography so its possible I messed something up a setting there.

Thank you!

Kyle Messina    

[Mathew McLaren]

Hi Kyle,

Did you try deconvolving the tomograms with the --deconv option when reconstructing? The standard Aretomo output which uses the simultaneous algebraic reconstruction (SART) approach which gives it a lot of contrast. The tomograms that come straight out of Warp are CTF corrected and can sometimes have poor contrast as a result, but the deconvolution (or denoising if you want to try that) can improve it quite significantly at the cost of removing some high frequency information.

Cheers,

Mat

--
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/9dad63e9-282a-41d1-971c-c87716765cd8n%40googlegroups.com.

[Alister Burt]

Hi Kyle,

Have you run through the tutorial before working with your own data to check that everything is working as expected in your system? (estimated time to complete: 2h)

The AreTomo wrapper works on most but not all tilt-series in the tutorial in my hands...

Are you sure you're setting --alignment_angpix correctly when importing the results? 

The alignment import routine used by both the IMOD and AreTomo wrappers is pretty well validated at this point which unfortunately suggests user error.

As you noted, you have to be careful when running AreTomo as various combinations of flags make it output IMOD files with very different contents... very frustrating - we recommend using AreTomo v1.3.4

Cheers,

Alister

--

[Alister Burt]

Mat might be right, it could also just be poor alignments and lack of filtering/binning... here is a slice of the unfiltered tutorial tomograms at 10Apx (raw data at 0.78Apx) for comparison.

Cheers,

Alister

To view this discussion visit https://groups.google.com/d/msgid/warp-em/CAA%3DxVZtXnbu11fuGt%2BE5f3q_%2Br6%3DKAKem18TvpKCvJj5uV%2Bh2w%40mail.gmail.com.

[Kyle Messina]

Hello,

Mat: I can definitely try running with the --deconv option to see if it helps this dataset, I'm never quite sure when to try options like these but I guess trying them out is the only way to figure it out.  I'm sure it'll be more obvious as I start working with more tomo datasets.

Alister: I have not run the tutorial dataset since I did not think it would matter (famous last words).  But problems like these are exactly why you suggest trying the tutorial dataset first to figure out if its program or user error (probably the latter in my case since I'm still learning).  I'll try Matt's suggestion first just to see if that helps out at all, and if that doesn't work I'll give the tutorial dataset a whirl to double check everything is running fine.

The Aretomo Wrapper could be user error but it looks pretty straightforward to run, especially compared to all the options Aretomo normally has.  To my knowledge, I have the --alignment_angpix with the correct pixel size and the import alignment routine seems to be working fine since it does see, and import all my Aretomo1/2 tilt series with no obvious issues.  Also thanks for the version recommendation since I found out not to use Aretomo2 but never wrote down the recommended version.  Also that's the kind of contrast I'm looking for! It doesn't need to be super dark or high contrast, I just want to be able to see particles/membranes without struggling.  Once I run through everything, I'll repost what the issue and solution was (hopefully).

Thanks!

Kyle

[Kyle Messina]

Hello,

I finally had a chance to run through the practice dataset for WarpTools and I had no issues getting a reconstruction that looks identical to the one Alister showed above. So for my other practice dataset, its almost certainly user error and not the program resulting in issues (not surprising).  I'll start fresh and take note of exactly what I do and post it here.  It's probably something small I'm not catching in my commands or a mistake I'm not aware of causing the issue. 

From,

Kyle 

[Alister Burt]

Hi Kyle,

Just to be clear, this doesn't mean user error - it likely means AreTomo is performing poorly for this data.

Cheers,

Alister

To view this discussion visit https://groups.google.com/d/msgid/warp-em/acb86b11-bb43-4c6e-93e5-d36fd4bb8b16n%40googlegroups.com.

[Kyle Messina]

Hi Alister,

That's the part that seems odd to me since when I run Aretomo independently from WarpTools on the other (non-WarpTools practice) dataset, the resulting Aretomo reconstructions do look good.  I wondered if the Aretomo wrapper had an issue but on the WarpTools practice dataset, it works great and looks like the IMOD reconstruction.  Are there any things too look out for when using the Wrappers or when using Aretomo separately and reimporting the data into WarpTools that I may have missed (i.e. like tiltcor should be -1 or 0)? 

The good news is that I just collected another tomography dataset yesterday so I can try using that data set on Monday (cell membranes/actin/membrane bound proteins).  I know all the collection parameters so I know that won't be a possible confounding factor. I will probably stop short of subtomo averaging since its a small test dataset but I can at least use it for testing up to reconstruction and see if it is a dataset or collection parameter issue (i.e. I'm using the wrong values). I'll probably just use the WarpTools dataset for testing out subtomo averaging since its a nice collection of particles.

Thanks again for all the help!

Kyle

[Kyle Messina]

Hello again,

Using the practice dataset, I was able to get through the entire tutorial with almost no issues and currently Relion is performing 3D refinement on the exported particles.  I'll go back to that other dataset I was struggling with sometime in the future.  In the meantime I started working up the dataset I collected on cell membranes and had a couple questions and an issue.  

Questions: 

1. To make sure I'm reading it correctly, when I need to provide the exposure value does WarpTools want the total dosage (not rate) of each movie? 

2. For the tilt series settings file, how do you determine what the values of X and Y should be for tomogram dimensions? I noticed for the practice dataset, the movies are 5760 by 4092 but the dimensions for the settings file are 4400 by 6000.  Is there a particular reason why you deviate from the original values? Also, does the z-value in the settings file matter when using the Etomo and Aretomo wrappers (aka does the aretomo --alignz need to match the z-value in the settings file)? 

3. When you use the Aretomo wrapper, for --alignz do you give the unbinned value in Angstroms and then it is auto converted to the correct binned pixel value? Or do you give the binned value in Angstrom and then its auto converted to binned pixels?

4. If you provide a bin value in the frame or tilt settings file, is that binning value applied to the processed frames and/or tilt series?

Issue:

I decided to take my personal dataset for a spin in Warp and everything seemed to be going well until I hit reconstruction (end goal).   All the resulting Warp reconstructions looked like the provided image (Bad Reconstruction) so i went back to check if the alignments were off.  Oddly enough, all the alignments looked great but I noticed that the features were distorted towards the edge of the micrograph. In the provided image (Odd alignment) you can see the hole is not perfectly round and it has a weird curve at the top and the left side. Depending on the tilt, the both curves will move.  Any idea if this is causing my poor reconstructions or what it might be from? I used the Aretomo Wrapper for these alignments if that matters.  I can try to use the Etomo patch alignment tomorrow but I've never seen it have good performance in low detail micrographs (could be me missing something).

-Kyle

[Alister Burt]

Hi Kyle,

Responses inline below

Questions: 

1. To make sure I'm reading it correctly, when I need to provide the exposure value does WarpTools want the total dosage (not rate) of each movie? 

Yep, total dose per movie. Can also be provided as dose per frame if desired by providing a negative value.

 

2. For the tilt series settings file, how do you determine what the values of X and Y should be for tomogram dimensions? I noticed for the practice dataset, the movies are 5760 by 4092

but the dimensions for the settings file are 4400 by 6000.  Is there a particular reason why you deviate from the original values?

Yeah this is a slightly strange one, the tomogram coordinate system has the tilt axis of the images along its y axis. As a result, if your tilt axis angle (convention: ccw positive from y) was exactly 90 degrees then your y dim in the tomogram would need to match the x dim of the tilt image.

For the tutorial data, it's a little off 90 - the values were chosen to encompass the whole field of view

 

Also, does the z-value in the settings file matter when using the Etomo and Aretomo wrappers (aka does the aretomo --alignz need to match the z-value in the settings file)? 

No, these are independent

 

3. When you use the Aretomo wrapper, for --alignz do you give the unbinned value in Angstroms and then it is auto converted to the correct binned pixel value? Or do you give the binned value in Angstrom and then its auto converted to binned pixels?

from the helptext: "value for AreTomo's AlignZ parameter in Angstrom (will be auto-converted to binned pixels), i.e. the thickness of the reconstructed tomogram used for alignments"

 

4. If you provide a bin value in the frame or tilt settings file, is that binning value applied to the processed frames and/or tilt series?

binning specified there determines the pixel size in frame series output - this has downstream effects on tilt series processing, tomogram dimensions are specified in unbinned pixel size relative to the binned pixel size (from memory)

so i went back to check if the alignments were off.  Oddly enough, all the alignments looked great but I noticed that the features were distorted towards the edge of the micrograph. In the provided image (Odd alignment) you can see the hole is not perfectly round and it has a weird curve at the top and the left side. Depending on the tilt, the both curves will move.  Any idea if this is causing my poor reconstructions or what it might be from?

How are you checking your alignments?

Tilt series alignment is one of the least robust steps in the pipeline, what you are seeing looks like the results of failed tilt series alignment

Cheers,

Alister

To view this discussion visit https://groups.google.com/d/msgid/warp-em/dc61b434-8a11-49fd-9cd6-3e5841654ca9n%40googlegroups.com.

[Kyle Messina]

Hi Alister,

Thanks for answering all my questions.  For Question 2, were the 4400 by 6000 determined by trial and error? For Question 4, I saw that description but was not entirely sure. I'm reading it as the value I provide will be convert to pixels and binned, is that correct? Otherwise, that all make sense!

For the tilt series alignment, I did a visual check of the *_aligned.mrc file in the tiltstack folder but if its as finicky as you say then I probably need a more quantitative approach to determine how good the alignment is.  I know from using etomo that a coarse alignment may look good to our eyes, the fine alignment helps a lot towards getting a good reconstruction.  I know IMOD gives a quantitative value for alignments (although I forget if that's saved to a file), does Aretomo have a similar output to check?

Thanks,

Kyle 

[Kyle Messina]

Hello,

Apparently it's Aretomo this time around causing the alignment issues as I've overestimated its ability to get alignments on these micrographs.  There's less details than I thought in these tilt series.  Thanks again for all the help!

From,

Kyle