Provide initial structure

trying to run Ab-initio allowing for 3 classes I get the error:

"File “/cryosparc/cryosparc-compute/”, line 162, in
assert False, “Meta data file is not of a recognized type.”
AssertionError: Meta data file is not of a recognized type.

so, do the classes have to be already defied in the star file or can cryosparc classify in the run?

In the GUI it says: “'The number of classes. Each class will be randomly initialized independently, unless an initial structure was provided, in which case each class will be a random variant of the initial structure”

So, if I understand correctly, it can do so but only when I provide an initial model. But, where can I enter the path to this initial structure?

Thanks a lot, Dieter

Hi Dieter,

I’m not sure how that error happened, but to answer your question:
No, we don’t read anything from the star file for classification - cryoSPARC classifies while reconstructing the 3 classes ab-initio. If you want to provide a reference structure (essentially making it non-abinitio classification) that is also possible. In 0.2.47 you would provide a path as the structure input, but in version 0.3 released last night you would simply select a structure from the experiments page by clicking the “Structure” button under the structure you want. That will show up in the “cart” on the right, then when you click New Experiment -> Ab-initio cryoSPARC will use the selected initial reference. If you want three classes, simply select the same initial reference three times (by clicking the “structure” button repeatedly) and that will instantiate three classes from the same reference. You can also instantiate different classes from different references.

Hope that helps,

And also if you want to import a reference structure from somewhere else there is now and “Import Density/Mask” task in the experiments page. But right now there may be issues with greyscale if you import a Relion or Frealign map - we’re working to fix this.

Thank you very much!

Sorry, but I still do not entirely understand. Can you please clarify:

  1. from a relion2 star file cryosparc only reads:

_rlnImageName #1
_rlnDefocusU #2
_rlnDefocusV #3
_rlnDefocusAngle #4

can other fields be present but are ignored?

needs “_rlnImageName #1” always refer to an image in a single image stack (e.g. made with relion_stack_create) or can it directly read the images from the directory structure of relion2? So far I always made a stack with all images.

  1. I can still not find the parameter specifying the number of classes that should be considered in the ab initio run. I see “Select Input Classes” and the “Class probability threshold”. Where do I specify the number of classes? Or does cryosparc decide on the number of classes based on the threshold above?

Hi Dieter,

Not sure re (1) (although yes, definitely other field can be present, I have all the usual fields in the star files I am using as input), but as to (2) in 0.3.1 there is a box on the right hand side of the Experiments tab where you can enter the number of ab initio classes, see attached pic. In you could enter it in the experiment setup tab, but that no longer seems to be the case.


Hi Oli,
thanks! Found it! Now I have to wait until the first run finishes since the ‘kill button’ has disappeared in the updated version…

Kill button should still be there! at least it is on mine - if you go to the results page where you launch it there should be Kill and Clear buttons


OK, I think restarting helped…

I can still not find where to enter the number of classes! There is this box at the right side, but whatever I put it is erased prior to the ab initio run…

Hmm, that is weird - that box works for me. Sounds like a bug perhaps specific to your system?


It worked! Problem is that in the Setup Section there is no indication of how many classes were selected. Class selection is outside the setup - as far as I can see -