Data collection using K3 and EPU

Hi everyone. I have a question regarding EPU. I would like to know how to decide upon the pixel size for data collection in a K3 camera? Previously i collected some data for 160 kDa membrane protein at 105k mag at 0.85 A, which never reached a resolution even around 2.5A (Nyquist).
Recently i attended a webinar by TFS and it was said that it’s preferable to do data collection at super resolution mode with 2x binning. I was wondering how 2x binning is better than collecting at normal counting mode with no binning. Do we need to tweak some parameter in motion correction to 2x binned data? Also, i would like to know which format of data collection is preferable, a mrc or a tiff format with tiff lzw non gain correction?
Thank you.

Hi @Das,

The K3 always acquires with super resolution, and bins the output if you select counting mode. IIRC, it bins in real space, not Fourier space? So saving in super res is the closest you’ll get to “raw” K3 output. Then in patch motion correction, Fourier crop. In CryoSPARC (and Live), this is “Output F-crop factor” in advanced (hidden) parameters during configuration, and using 0.5 would Fourier crop down to “counting” mode. In RELION, it’s the “Binning factor” in the motion correction job “Motion” tab… and it needs to be set to 2.

MRC files (if gain corrected) can be pretty large. LZW compressed TIFF files with a separate gain reference are more storage efficient, especially if saving in super res…



Hi @rbs_sci
Thanks for the answer. So for my previous collection where i collected in counting mode in tiff format using tif lzw non gain correction” do i need to use 0.5 f crop factor in patch motion correction?
Also i would like to know what is the typical size of one movie collected in SR bin2 format. My one movie in counting mode “tiff lzw ng corrected are around 250 mb (0.85 pix, 40 frames).
Many thanks.

If it collected in counting mode, it is already at physical pixel size and you shouldn’t use F-crop 0.5 (unless you wish to bin the data more).

Collecting in Counted more or in Superresolution + bin 2 should result in about the same file size. The difference is EPU does the binning in the first setup, while GMS/DM does the binning in the second setup. In both setups, TFS recommends you collect the data as gain corrected, as the K3 gain ref file is in superresolution. If you want to collect movies without gain correction, TFS recommends you collect Superresolution bin 1.

How do you currently gain correct you counted movies? By converting and binning the gain ref file from GMS?


Hi @boggild,
Thanks for clearing the doubt. I collected two sets. The first one in counted mode 0.85 pix no binning. Non gain corrected lzw tiff.
While processing, i import the tiff files and give the gain reference file( .mrc) generated automatically in the collection folder. Is this correct?
The second data was in counted mode but gain corrected. So during import i did not use any gain ref file.

Yes. :slight_smile: Although you might have to rotate and/or flip the gain reference. The imported movies will likely look obviously bad if the gain needs flipping or rotating 180 degrees (and import will fail if it needs rotating 90 or 270 degrees complaining about dimensions).

1 Like

To add to the other responses, logistically, setting EPU to collect “super resolution/bin 2” provides up to twice the throughput vs “counted/bin 1” because of where the binning is occuring (GMS vs EPU, respectively). At least it used to be the case. It probably still is. However, unless I’m mistaken, I believe this is binning and not fourier cropping as one might benefit from with the SerialEMCCD plugin, so there is a DQE penalty to consider with respect to raw super resolution.


1 Like

@boggild @rbs_sci @leetleyang
So, in a nutshell for K3 using EPU either counted mode with no binning or super resolution with x2 binning is recommended. The later is faster in terms of acquisition. And if i collect at SR x2 binning then, in cryosparc patch motion correction we need to give 0.5 F crop factor. Also we can use the K3 gain (.mrc file) directly in Cryosparc, but it may need flip/rotation. So did i get it correctly?
Many thanks

I wouldn’t recommend counted bin1 with the K3. I mainly use super res with bin 2 enabled in EPU with gain corrected movies in compressed tif, but if you prefer slower data collection speed with better data quality, you could collect in super res bin 1 without gain correcting the movies (still compressed tiff to save space) and do the gain correction in Cryosparc with F-crop = 0.5.

If you collect SR bin2 you should not use 0.5 F-crop (unless you want to bin further), as the output is already at physical px size.

Take a look at Dustin’s SciLifeLab post, linked in above and here, for details (and the discussion in the comments below it).

Since I usually collect using gain corrected output it might no longer be relevant, but it used to be that the gain ref file output by EPU wasn’t valid for K3 data. Instead we had to copy the gain ref directly from GMS (the “.m1.” one). You probably need to convert the file format and flip/rotate depending on format of gain file and movies.

1 Like

@rbs_sci @boggild @leetleyang, I have collected a data set using the super res bin/2 at 105kx mag (k3/300 kV). I see that the images are 2kx2k instead of 4kx4k that we see in counted this fine?
Also, the pixel size here is 1.7 A. So now when i process in cryoSparc,should i give 1.7 as the pixel or 1.7/2 as its bin 2? What all parameters i need to change in patch motion correction/motion cor2?
Many thanks.

If collected as super res bin 2, it should not output 2k movies but instead 4k (binned from the 8k super res). For our K3 Krios, the pixel size from 105kx super res bin 2 is 0.83 Å/px. Are you sure it wasn’t set to Counted with bin 2?

For motion correction, you should input the true pixel size of the data, so if it was binned once more (making the data 1.7 Å/px at 2k) you should input 1.7 Å/px.

@boggild Finally i could solve the problem. Now the microscope has given a .dm4 gain reference and cryosparc needs an mrc. Could you suggest how to do that? I found that IMOD dm2mrc can be used. I tried to use the online version ( - Resources: IMOD online, but it does not work.
Many thanks

I would use IMOD, but have it installed locally. I haven’t tried that online tool.

How did you set up your data collections now - are you using non-gain normalized tiff movies in super resolution with no binning?

Hi @boggild.
I installed IMOD and converted the K3-Gain Ref. x1.m1.kv[300].dm4 to mrc, by “dm2mrc” command. After that, I used the command “newstack -in gain. mrc -ou gainbin2.mrc -bi 2” as the data was collected in Counted SR bin2. The microscopist suggested a horizontal flip for the gain, which I applied during the import movies in cryosparc by “Flip gain ref & defect file in X= YES”
Have I done it correctly? kindly suggest.
Thank you.

We usually collect gain corrected movies, so I don’t recall the correct flip. I suggest you try both X and Y flip on just a few movies and see if where gain pattern is effectively corrected. You can also try without any gain file to see how it looks for reference.