We are seeing some unusual patch motion correction curves and were hoping to get some advise on how to mitigate the issue. We collected K3 movies on a Titan Krios with an object level pixel size of 1.06Å/pixel and a dose rate of 1.2 electrons/Å2/frame for a total dose of 60 electrons/Å2. During patch motion correction we see a region of the micrograph having fairly large erratic motion. On inspecting the motion corrected micrograph we see the region that showed large motion in the patch motion correction plot looks less clear/sharp than other areas (red square in the attached micrograph screenshot). I am attaching a representative case. On comparing a simple summation of the frames without alignment with the aligned micrograph we see that motion correction is working and even the area where local motion is large becomes better defined compared to the simple sum of frames. Also, in the simple sum of frames some regions do appear blurrier than other regions and that tallies well with the regions showing exaggerated local motion.
We have two questions: What might cause such exaggerated local motion? and What can we do got those regions align better?
We have ruled out a few possibilities that we could think of.
It is not an issue specifically with patch motion correction and frame alignment with motioncorr2 (5X5 patch) leads to similar issues.
It is not an issue with incorrect gain since with/without gain did not make any difference.
3)Removing the first 5 frames or last 10 frames did not make any difference.
We are still processing the data and are not sure if this affect 2D classification but I would assume that it will given that some regions of micrographs are visibly less resolved than others.
When I see large motion like that, and the micrograph looks like you display, my instant thought is “ice is very, very thin”.
Although give the horizontal line in the middle of your micrograph, check you gain reference, which if poor can impact motion correction due to artefact correlations.
Thank you for the response. It is possible that the ice is a bit too thin. The protein is relatively small (70 kD) and we used detergent to get the particle go into the holes. Is there any way to better align the movies in such cases?
There is a very strong line right through the middle of the detector chip and that never gets corrected. Perhaps I should ask for a defect file from the facility. However, there is low chance of the gain is causing any issues since the large motion area is not in a fixed patch for all micrographs.
You collected two images per hole, and the beam of the first image overlaps the camera from the second, so you’ve double-exposed those particles by the time you collect their data. Would expect this happens for ~every other image.
Gain ref and hardware dark ref should get rid of the line from your K2, no need for defect file. gains must be collected with the correct dose for the day’s data collection (not true of Falcon eer).
There are no particles visible in the high motion area, which would support the idea of extremely thin ice.
Not really. You could try whole frame motion correction to see whether there might be particles there or not, but it won’t help - just act as a diagnostic.
That strong line means your gain/dark references need updating. Defect files mark bad pixels (either permanently on or permanently off). Gain reference/dark reference corrects for the fact that each functional pixel has a slightly different response to excitation. This is not a large motion issue directly, but a more systemic issue with your camera/data/processing. I was offering something else to check, since I have seen extremely poor motion correction output when the gain reference is bad.
Well, one would hope that the user (if trained and acquiring data themselves) or facility staff would understand that overlapping beams so that a camera exposure area is double exposed would be a bad idea…
Absolutely. OP states K3, but image is not K3 ratio. And the gain stability of the Falcon cameras completely blows the K2/K3 out of the water, it’s not just an EER thing.
Thanks for pointing out the possibility of double exposure. I double checked when setting up the queue that the image areas are not exposed twice. Also, as you pointed out if this is indeed an issue with double exposure then we should see the issue in every alternate image. But that isn’t the case.
Dear rbs_sci
Your point is well taken. The gain is clearly not doing its job. The frames were saved as tiff files. The gain was in dm4 format and I converted it to mrc using dm2mrc command and then binned by 2. I did not rotate the gain in any way because I was under the impression that the way cryosparc treats the x and y axis of a tiff image do not require any rotation of the dm4 gain. But perhaps some sort of rotation is needed after all? I will investigate today.
Forgot to add one more thing. @rbs_sci is absolutely right that the micrograph screenshot is not of a K3 image ration. I posted the screenshot of part of the micrograph. Sorry about that confusion.
I usually generate a gain file from the micrographs themselves in cases where the provided gain isn’t very good. Usually sum_all_tiff_files from cisTEM or relion_estimate_gain work great.
I will give both cisTEM and Relion a go today. I was wondering if you would know whether any rotation of the Relion/cisTEM generated gain would be needed for doing patch motion correction on tiff frames in cryosparc.
