Leginon

Someone is collecting data now with 10 second (200 ms/frame) counting movies with 4 image-shift exposure targets plus one focus target from one hl image. It is done in 176-191 seconds, so that makes it 44-47 seconds per image.

In the parenthesis I give you possible tweak to maximize the throughput on the step.

This includes

• acquiring hl image: 13 seconds total with 2.5 s pause before imaging.
• automated target finding: 1 second (This is a big time saving. Without queuing, this means it is o.k. not to transform targets since the targets will be processed right away and drift is minimal).
• focusing: 28 seconds with 2 beam-tilt autofocus, once per hl image targets without drift monitoring (If you have a very flat grid, you can skip this more often. Drift monitor is also fairly time consuming. If you can avoid it, given a good stage and grid prep, you can skip drift measurement like in this case).
• preset and target image shift settings: 3 seconds (Arctica lens normalization should work ok so that you can turn off cycling).
• pause before imaging: 5 seconds (This can be reduced to 3 seconds in image-shift targeting on a good stage).
• the first image is giving extra imaging time: 10 seconds
• k2 imaging and transfering back the sum image: 20 seconds (This can be reduced with early return configuration in demsem.cfg (3.2), but is only advantageous if you are moving to targets mainly by stage shift since in image shift case the next movie is taken so soon that it still has to wait for save to complete).
• saving the image and meta data: 1 second
• display the image: no image displayed = 0 second (This is the place you can save time easily. Just disable Display Image in the advanced setting).

The more final images you can take without moving the stage, the faster the data collection will be.

Hi Anchi,

Thank you for such a detailed description! 4 exposures per hl image would be great, but I'm afraid our stage movement may not be precise enough. I really struggled to get a modeled stage position to be 'OK', so we'd have to redo the modeled stage position to be able to move more precisely.

With the new stages on Krios and Arctica microscopes, do you have any tips for creating the modeled stage position? Could I trouble you for a screenshot of one of the stage models?

Mike,

When I say hl image, it does not mean that it is one hole. Typically, this kind of speedy experiment uses 1.2/1.3 C-flat or equivalent quantifoil. In each of the hl image, it would see 4 holes, and each exposure is taken at the center of the holes with the beam cover the hole to touch the support foil, may it be carbon or gold.

Unless you are going for sub 3 Å resolution, you don't need to use stage move to get to a target within 2 um radius on a Krios. We have done that plenty of times. Arctica is likely o.k., too. You need to confirm that with your own experiment, though. See the procedure to measure beam tilt induced by beam-image shift in

Cheng A., Tan Y. Z., Dandey V. P., Potter, C.S., Carragher, B., 2016 Strategies for Automated CryoEM Data Collection Using Dierect Detectors. Methods Enzymol. 579:87-102

Regarding the modeled stage position, My experience is that Krios and Arctica don't benefit from higher harmonic terms any more. I am planing to phase it out, at least no longer used as default. When we do use stage model, I fit it to 0 terms, effectively make it a straight line fit and give very similar behavior as a 2x2 matrix calibration.

There are two strategies in moving to a target. First, use automated target finder and depth-first traversal, and the Second, manual targeting and breadth-first traversal. See Queuing option. Depth-first requires smaller movement, and is in general more accurate. We can do a target of 0.4 um image length on a 1.2 um hole without touching the support film on newer Krios or a well-maintained tecnai (presumably Arctica) stage. With breadth-first queuing, target adjustment to compensate the returning error is usually needed as people using this tend to need very accurate targeting, such as targeting filament.

There is a bit of variation in the need of backlash correction among different generation of the FEI scopes, though, even within the same scope model. You can experiment it. There is a button in Navigation node to do a test. Beaically, it moves from the current position out by some distance you defined, pause for a short time and then come back. At each return, an image is taken to see how reproducible it is. You can try it with and without stage backlash correction (change in pyscope/tecnai.py. and look for self.correctedstage definition at the beginning of the class ).

By the way, I heard that the current record is 100 final exposure per hour now. The grid was oriented in a way that 5 exposure targets were possible in the one 6 um length hl image.