performance of unblur

Submitted by Antje on Fri, 2015-10-16 05:50

Forums

Unblur & Summovie

Hello,

we are currently trying the unblur dose filtering on a 20S test dataset, acquired on a Falcon 3 with a total dose of approx. 70 e A^-2 over 2 minutes.
The frames have been aligned with another program and I used the aligned frames with a dose of 2 e A^-2 per frame for the dose filtering procedure using summovie and handing over shift files with zeros in all shift entries.

To evaluate the dose filtering I compared the refinements done in RELION of the following three datasets:
a) particles extracted from aligned sums
b) particles extracted from aligned sums and polished in RELION
c) particles extracted from aligned and then dose filtered sums (using summovie)

Unfortunately, the resolution for the dose filtered reconstruction (c) was worse than the resolution of the other two. Now, I am wondering, where I might have gone wrong.

This is my script that loops over all the movie stacks:
--------------------------------------------
#!/bin/bash

for i in "$@"
do
/fs/path_to_summovie/sum_movie_openmp_7_17_15.exe << eof
$i
29
${i%_movie.mrc}.mrc
${i%_movie.mrc}_shifts.txt
${i%_movie.mrc}_frc.txt
1
29
1.4
YES
2
300
0
YES
eof
done
--------------------------------------------

Something that puzzles me:
At the beginning I accidentally ran the script above with an erroneous dose per frame of 1 e A^-2 and extracted the particles from those, which let to a refinement result with a resolution better than the refined summed particles (a), but still worse than the polished particles (b).

I would be very grateful for any ideas what might have gone wrong!
Antje

timgrant Fri, 2015-10-16 08:58

Hi Antje,

If you have a pixel size of 1.4A, your exposure per square angstrom per frame was 2 electrons and you're using a 300 kV machine these options seem correct.

Maybe you can provide me with a bit more information :-

1. How much did the resolutions differ?
2. The refinement performed with particles extracted from the aligned sums, was that using all the frames, or just some of them?
2. You say approx. 70 e/A2, but the in sum movie you're using 58 e/A2, is this correct?
3. How certain are you that your dose is correct? At 4 seconds per frame, I assume this is doing counting - are you taking the coincidence loss into account?

Thanks,

Tim

Antje Mon, 2015-10-19 09:08

Hi Tim,

I meant to write that the total exposure time was 2 seconds, not 2 minutes. Very sorry about the confusion!
Anyways, here is the reply to your questions:

1. The resolution values for our test dataset were the following:
-> particles extracted from aligned sums: 4.4 A
-> particles extracted from aligned and exposure filtered sums: 4.65 A
-> particles polished by RELION: 3.9 A
2. For all extracted particles the same number of frames contributed. We discarded the first frame as well as the last few (see below).
3. & 4.
I wrote approx.70 e/(A^2) because we acquried the data with a nominal dosis rate of 35 e/(A^2 s), determined using the flew screen. Of cause all dosis values are just as exact as our flew screen measurement. The per frame dosis, doesn't add up to this number, because we deleted a few frames during the frame alignement procedure.
The details are: We acquired the images on a Falcon3 NOT in the counting but the "capture" mode with approx. 32 frames per second. The number of frames sometimes varies slightly (+/-2 maximum) from stack to stack, which we were told by FEI is normal. For this reason we only used the first 60 frames to have the same number of frames in each stack for the processing. We also deleted the first frame, so I should have set the pre-exposure to 1 e/A^2. For disk space reasons, we always added up two aligned neigboring frames and therefore end up with 29 frames in our aligned stacks.

Do you see something that might cause the problem ?
I assume the missing pre-exposure of 1 e/A^2 doesn't explain the decrease in resolution from the sum to the filtered sum ?

If I remember correctly, the dosis measurement when using a K2 is more precise, because you don't have to rely on the flew screen. Might it be, that the dosis determination using the flew screen is not precise enough for the unblur procedure?

Thank you very much for your help!
Antje

Axel Mon, 2015-10-19 12:10

Hi Antje,

I would agree with your guess that the dose rate calibration is off (perhaps as much as two-fold). The flew screen method you describe is prone to errors depending on how well the screen itself is calibrated, and how close you are to the dose as which the screen was calibrated (the sceen has a non-linear response to dose).

You could try a number of different dose rates in summovie, calculate reconstructions, and see where you get the best reconsctruction. You may even be able to fit the maximum in the resolution to get you dose rate.
HTH,

Axel

Antje Mon, 2015-10-19 12:26

Hi Axel,

thank you for your reply.
Yes, we figured that this would be something we could try...

Is there someone, who has tried the summovie dosis filtering on data, for which the dosis had been determined by the flew screen method and who would share his /her experience?

Many thanks in advance,
Antje

timgrant Mon, 2015-10-19 13:50

Hi Antje,

I think that Axel is most likely correct. The fact that you got a worse resolution with the exposure filter than using all of the frames strongly suggests that your total exposure estimate is wrong. As Axel suggests you could try different values to find the optimum. If you do this, you may well be able to use this conversion factor for future datasets collected in a similar way.

Thanks,

Tim