I have a data set collected at 25° tilt in superresolution mode. I am processing it with cryoSPARC 4.2.1. When the micrographs were binned twofold in “Patch Motion Correction,” the following “Patch CTF Estimation” gave an average tilt angle of 23.7°, which is reasonable. However, when the micrographs were not binned in Patch Motion Correction, the Patch CTF gave an average tilt angle of 12.3°, which is close to half of the angle of the stage. Comparing the results of these two approaches, all statistics are comparable except the average intensity, which is ~4 times higher in the binned data and makes total sense.
What caused the tilt angle to be halved? Any suggestions? Thanks!
Maybe a small number of (likely bad) micrographs with very different outlier values?
GPU computation is non-deterministic, there are going to be some differences like this just by rerunning the same job with e.g. a slightly different resolution limit.
Thanks for the input. I am sorry I didn’t state that this is not a small data set (~3500 movies). There are some outliers, but I don’t think those made much of a difference.
Here is the tilt angle histogram from unbinned data.
I tested on a different 40° tilt data set and got the same result. Unbinned data gave an average tilt angle of half of the stage tilt, while the twofold binned data gave an average tilt angle close to the stage tilt. I am using cryoSPARC 4.2.1 as well.
BTW this calculated tilt isn’t going to be used for anything elsewhere in processing - it’s only relevant if you wanted to threshold micrographs by the estimated tilt (which I would recommend against doing anyway - it should still be calcuated correctly though!).
Yes, I agree with you. I don’t think the tilt angle is used anywhere else. However, I was rather surprised when I first saw it in cryoSPARC Live, and it made me double-check the stage angle in data collection. If it is a bug, it should be fixed. Otherwise, people could be thrown off by that.
Just to add @hsnyder - we see the same thing in v4.4. We are collecting a dataset with 40 degree sample tilt right now, and the value reported after Patch CTF (excluding outliers) is almost exactly 20degrees. This is for data processed in CS Live, v4.4. Was there any further info on this issue? Is the underlying defocus estimation wrong, or is it purely an incorrect calculation of the tilt?
Would be good to compare an example mic to CTFFIND5 to see how (a) the estimated tilt compares & (b) how the estimated defocus compares at a given off-axis position
Also I wish there was a way to import micrographs with CTF estimates to facilitate such comparisons… support for micrograph star file import would still be a welcome addition!
I think this is probably not the problem, but can you try override knots with a low number (2-4) in patch CTF with more knots perpendicular to the tilt axis, so 2x3, 3x4 or vice versa?
It was worth a punt, but with testing a 20 mic subset there was no change - approx 15 degree tilt regardless of what knot settings I used (tried x=2 - 6, y=2-16).
…
Had time to quickly check two micrographs with CTFFIND5, where output is:
Sorry, no news yet. We haven’t forgotten, we’ve just been busy with other features. That said, thanks for bringing this up again. While I can’t promise a specific timeline, it’s helpful to know what issues are causing recurring headaches for users.
@rbs_sci did you try this with a corresponding particle set? Just wondering if the estimated tilt angle means anything in CS - does it correspond to significantly different per-particle defocus estimates, or is it just a miscalculation of some sort?
It’s a purely aesthetic error; it appears to have no impact on further processing.
I got that dataset - combined with an untilted dataset off the same grid - to 1.9 Ang, found some other interesting things they weren’t expecting; now just waiting for collaborators to move forward with the rest of the project… independently the two datasets got to 2.1 (tilted) and 2.2 but a bit anisotropic (for the untilted).
Plotting angles from the orientation distribution shows the alignments are more the angle expected than the one reported from the SR CTF tilt estimates, and comparing projections of the 3D vs. 2D they look dead on for 30 degree tilt projection.
