Irregular shifts on holey substrates

[asarnow]

Hi Fabian,
After going back to some non-lamella imaging, I started using PACEtomo for e.g. imaging cell peripheries with irregularly spaced targets, that always have stable focus positions nearby. If you don't mind I wanted to sanity check what I'm doing with you.

I'm using targetsFromMontage with View montage maps. The high-defocus calibrations were done and View/Record alignment is pretty excellent all around the shift range at -50µm defocus offset, so I would like to rely on View alignments as much as possible. I used refineGeo = True, addAF = True, previewAli = False, viewAli = True, with the geometry left at 0˚ pretilt and rotation.

I also tried some SPA-style imaging, with AddAF, previewAli = False, viewAli = False and with minTilt = maxTilt = 0 and noZeroRecAli = True. AFAIK using PACE like this is the most convenient way to collect at arbitrary shifts in SerialEM, sort of like it works in Leginon, useful for lacy grids or e.g. filaments or vesicles that are not all over the holes and can be seen in View.

• There still seems to be a Preview of the tracking target taken at the start tilt, which I can't figure out. I have realignToItem = True so it shouldn't be the one in realignTo().
• For untilted imaging, could there be an advanced option to skip finding the eucentric height? (Like this).
• For untilted imaging shift accuracy is improved using viewAli = True, but switching the low dose states while setting up the targets takes a large fraction of the time - can we possibly skip the GoToLowDoseArea("R") at line 1417 in the target setup loop? There is an error restoring the beam tilt unless I include the call...but moving it out of the loop seems to work...am missing something import here? (Like this).

[eisfabian]

Hi Daniel,

Thanks for playing with this!
There are a few things:

• I recommend using SPACEtomo for this. targetsFromMontage never really made it out of beta and SPACEtomo does the same thing but better. You can run the target selection on already manually collected montage maps.
• refineGeo is probably not useful here, it uses CTF estimation for geometry refinement.
• If you set addAF = True, make sure to pick a tracking target that is not actually a region of interest, because it will get burned.
• SPA imaging: I heard of someone wanting to use PACEtomo like this before. It might be worth making a much simpler script that can do that, I will look into it. This would of course incorporate your changes.
• SerialEM's realignToItem will also take a Preview image during the realignment routine, if the tgt reference is a Preview image (real or simulated).

Let me know if I missed anything or if you noticed anything else!
I will try to have another look at it next week!

Thanks again and best wishes,
Fabian

[asarnow]

Thanks for the reply! I will try SPACEtomo...couple more points/questions below.

• With addAF I was setting the focus position for the tracking target to be on the carbon nearby - in that case I thought the actual target wouldn't be exposed?
• To avoid the RealignToItem Preview shot, do you think skipping the tgtfile Preview alignment (when previewAli = False, viewAli = True) in PACE's own realignTo code will cause any problems? The View/Record alignment at 0 IS really is perfect for me.
• A simplified SPA script would be great! Based on my experience the remaining shift inaccuracies I see are caused by actual scope error. In other words, if I do a large mouse (image) shift, then the View will not quite be where I shifted. If I repeat it with an additional small shift, so the shifted View is on the target, then the following Preview/Record will be just right. So for SPA one would want to shift, realign (in View), and then record each target sequentially, instead of saving target shifts in a batch and then imaging in batch as PACEtomo does now for tilt series. Unfortunately I don't think there is a way to avoid many Low Dose Area changes if exact targeting is required. On the other hand If these intrinsic shift errors are OK then only the tracking target needs to be realigned.

[asarnow]

Fabian,
I learned about UseNavGroupForMultishot 2 and AdjustMultishotPattern 1 which enables irregular AFIS with minimal scripting in SerialEM. The primary limitations are 1) the calibrations, View/Record alignment, and the Multishot View/Record hole transform ("Step to and Adjust IS") must be very accurate and 2) accuracy is lost for targets on a different montage piece from the center position.

I think a couple of people are asking David for better support for these functions from the Acquire at Items dialog so it is probably not worth working on a SPA script until we figure out exactly what that looks like. One can look ahead and see that a nice feature might be to only do an image shift realign for targets on a different montage piece though.

Best,
-da

[asarnow]

Hey Fabian,
After doing some data collection based on the above, I have an actual PACEtomo feature request: would it be possible to use the multishot transform adjustment instead of other pre-acquisition target alignment?

I found it was very accurate, I would be happy having that level of accuracy based only on realigning the tracking target. (That is, no viewAli, no previewAli - just realign the tracking target and go, but using the multishot transform to the Record mag).

[eisfabian]

Hi Daniel,

Cool to hear you found a good approach for arbitrary SPA collection.

In principle, it should be possible if I can read out the image shifts somehow. I will have to look into it.
If that's possible, it's simply a matter of replacing the image shift values (in specimen coordinates) in the _tgts.txt file and turning all Ali settings off.

I will let you know when I had some time to check it out!

Thanks and best,
Fabian

[asarnow]

Adding this note for any later work. (Hopefully I can get to a minimal PR before too long).

The transformation is stored in a line in the settings file, for example
HoleAdjustXform 20 0 0 20 34 1.009630 0.005789 -0.002779 1.021203
which is an adjustment from mag index 20 to mag index 34 using the transformation matrix given by the last four numbers.