Incomplete ab initio solution?

Hi all,

I am using cryosparc 3.2.0 to get the structure of a protein complex (composed of two different proteins) and has difficulties to make sense of the map I got, although the resolution looks quite promising.

I initially selected around 875,000 particles from 2D classification, then an ab initio run with 2 classes, following a hetero refinement and I then run refinement with the best looking class (around 640,000 particles) which gave a map around 3.0A (applying C1 symmetry). However, the map looks a bit odd from one side (It has a sort of bump). Considering the class of this protein family, I expect a symmetric solution. Indeed, when I look at some of the images from 2D classification, I could really count obvious corners. I also tried to impose symmetry (such as C3, C6, D6) during hetero refinement and refinement steps, yet it resulted in worse resolution (>5.5A). In addition, I run ab initio job using the best class above (640k) and tried to divide into more classes (3 different) and run hetero refinement, but the solutions I got is more or less the same. Finally, I also picked less amount of particles from 2D classification which looked quite clear, but again one side the complex looks a bit unexpected to me


I am relatively new to cryo-em. So, I will highly appreciate if anyone could give any insights what might be wrong or what I am missing. Perhaps,

  • Not enough views (info) for this side of the complex, so couldn’t construct that part?

  • Symmetry problem? (I highly would doubt, I tried to impose all the available symmetry options)

  • Has to tune ab inito reconstruction parameters? (I run several runs by changing initial and final minibatch size (increasing)) and initial learning rate.)

  • Perhaps, it is what it is!

Thank you very much!


Hi Esra,

  • Do you see clear secondary structural features in your maps? At this resolution, if the reconstruction is correct, you should see clearly defined alpha helices and beta strands with identifiable side chains. Do you see this?

  • What does your orientation distribution plot look like for this refinement?


Hi Oli,

Yes, I see secondary structural features (alpha helices).

My orientation distribution plot looks like this:


This looks fine - there is no evidence of a particularly unusual or preferred orientation distribution here. If you can see alpha helices with clearly interpretable side chain densities, I would say the solution is likely correct.

for what it’s worth I think the images you are showing of the refinement look exactly like the 2D classes we are seeing. what is the bump you don’t expect? your first 2D class tells the whole story, there are 5 or 6 subunits of some thing in a spiral, something is bound to the outside of 3 subunits, there is flexibility (that’s why they are fuzzier) in the 5th/6th subunit(s) and they spiral like walking down a staircase. In any case, I don’t think there is pure symmetry here, maybe only pseudosymmetry. You can “make” a desired map (either you have something you were expecting or you can use chimera to make a fake map from a symmetric atomic model) and generate 2D projections then check if they look like your 2D classes. Try new Non-uniform refinement, it’s slower but will give amazing results.


yes exactly - looks similar to an AAA ATPase of some sort, they have a tendency to do that (shifting from a closed hexamer to an open spiral)

You can make projections of the map in equally spaced orientations using the “Create Templates” tool, and compare them to your 2D classes

1 Like

Hi again,

Thank you for the valuable comments and suggestions! Yes, it seems quite dynamic. I think I can move forward from here.