Subtomogram Reconstruction - Striping Artifact

[B Wimmer]

Dear Mailing List,

I have a question regarding a striping artifact I am seeing in subtomograms reconstructed using Warp (1.0.9 nightly 201104).

I acquired tomograms on FIB lamellae at 1.372 Apix. After particle picking outside Warp, I reconstructed about 150k subtomograms at 5Apix, 48px box size, 100A normalisation radius.

As an initial sanity check, I ran relion_reconstruct --ctf --3d_rot --maxres 20 on the resulting star file. In this sum of subtomograms (no angles provided), I see a very strong striping pattern in the YZ and XZ views (XYZ view attached).

Do any of you observe this? If yes, do you just ignore it or is it resulting from some incorrect parameters?

Full tomogram reconstruction looks fine and ribosomes from the same dataset go to about 12A from 8k particles.

Appreciate any suggestions and advice.

Best,

Benedikt

[Pranav Shah]

Hi,

The angles are required to insert the images in fourier space in the right orientation. In the absence of this information, the observed effect isn't unexpected.

--
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/fa00c42f-af75-4b9f-8c62-6154f410715fn%40googlegroups.com.

[B Wimmer]

Hi Pranav,

 

Thanks for the quick reply! I agree that in absence of angles, the sum of subtomograms should have a very visible missing wedge artifact. I would have expected the resulting volume to look like a noisy potato elongated along Z or something like that.

What I am worried about here is that there seems to be an artifact which is not noisy (would average out), but stripey, and which seems to be conserved across subtomograms (and thus retained even with 150k subvolumes), if that makes sense. The issue is that this pattern seems to strongly influence Relion 3.1.4 Refine3D and throws off alignments.

Best,

Benedikt

[Ricardo Righetto]

Hi Benedikt,

I'm not sure I understand what you mean. You used relion_reconstruct with the --3d_rot option but then you say "no angles provided". The --3d_rot option should exactly rotate the subtomos in the orientations provided in the STAR file to calculate the average. Or do you mean that the angles you have in the STAR file are all zero?

Besides clarifying that, I don't see anything blatantly wrong with this initial average. Perhaps the streaking is indeed related to the strong wedge effect and a small box size, but is not necessarily a problem IMO. How does it look after some Class3D/Refine3D in RELION?

Best wishes,

--
Ricardo Diogo Righetto

To view this discussion on the web visit https://groups.google.com/d/msgid/warp-em/5baf66de-a459-4061-ad26-87e15d1f94ddn%40googlegroups.com.

[B Wimmer]

Dear Ricardo,

Thanks for your feedback, I'll try to clarify below:

> I'm not sure I understand what you mean. You used relion_reconstruct with the --3d_rot option but then you say "no angles provided". The --3d_rot option should exactly rotate the subtomos in the orientations provided in the STAR file to calculate the average. Or do you mean that the angles you have in the STAR file are all zero?

The star file contains only the CoordinateXYZ + MicrographName + ImageName + CTF Image + Optics Group ID. You are right, the --3d_rot is redundant in this case - I just added it out of habit and the reconstruction without this flag looks the same.

> Besides clarifying that, I don't see anything blatantly wrong with this initial average. Perhaps the streaking is indeed related to the strong wedge effect and a small box size, but is not necessarily a problem IMO. How does it look after some Class3D/Refine3D in RELION?

Alright, already good to know that these streaks don't look obviously artifacty! I also only became aware of them as a default Relion 3.1.4 Refine3D job converges on the streaking pattern, rather than the blob (see image attached), so that's why I am trying to understand the root cause of this pattern.

Thanks again for your input,

Best,

Benedikt

[Jeff Songrong Qu]

Dear Benedikt,

Sorry to bother you and everyone on this thread after so long! I was wondering if you have a solution or explanation for these streaks? My recent RELION 3.1.4 job encountered a similar problem with radial streaks surrounding my target density. Thank you so much for any input!

Best,

Jeff Qu

[B Wimmer]

Dear Jeff, 

Ultimately, I don't think I was able to circumvent the issue with that particular dataset and approach. With the same tomograms, I was more succesful using a WarpTools > Relion 5 pipeline, particularly after projecting the subtomograms and cleaning them up by 2D classification and leaning heavily on the ab initio job in Relion 5. You can see my (ultimately unsuccesful) strategy in Figure S3 here. 

Hope that helps,

Benedikt

[Hamidreza Rahmani]

Dear all,

To me as well, this looks like an overfitting on particles that relion could not get a structure on. In my case I often need a better set of particles when I see this. 

Relion-5 on 2D projections from Warp has been much better at classification. 

Best.,

Hamid

To view this discussion visit https://groups.google.com/d/msgid/warp-em/a4789da8-3524-4a7d-a68b-f857912680a6n%40googlegroups.com.

[Jeff Songrong Qu]

Dear all,

Thanks again for the input! We have also been trying to use relion5/4 for the same particle set, and we observe the trend and were wondering if you have any inputs: 

1. In relion5/4, at finer sampling (in my case, 0.9 degrees, after 48A resolution), the structure started to disappear. Still, the noise becomes dominant in the solvent region (it should be densities at a different registry, but now it is completely in noise, and further refinement showed the noise became dominant). When I run relion_reconstruct, it does not show that pattern. I was wondering if there's a way to fix that issue, as I want to run classification on the noisy region. (the images are before finer sampling in auto fine and final refinement result)
2. In relion3, refinement on the same tomogram set at finer angles it showed the streaky noise in the pattern.

I was wondering if that could be resulted from either noise evaluation or improper processing steps, any inputs or further questions are really appreciated!

Best,

Jeff