NU refinement report 3.0 A resolution but the quality is bad

Hi everyone,
I am trying to process a Cryo-EM data (a small protein (90 Kd) in complex with RNA (30 Kd)), the 2D classification looks great and the NU refinement reports a resolution of 3.1 A, the overall map looks good (I could clearly see the RNA structure) but I found that the map quality is quite bad, the main chain is not connected and lost most details, please see below. It seems like the protein does not have an orientation issue, because we can reconstruct the map from relative low-resolution data and are able to dock the most domains in the map. Someone told me that it’s the refinement alignment issue for this high-resolution data, but could not figure it out. Does anyone have the same expertise, and does anyone know the possible solution? Really appreciate!
Screen Shot 2022-05-11 at 2.59.24 PM
Screen Shot 2022-05-11 at 2.59.09 PM


1 Like

Besides, the below are more information about NU refinement. Really hope someone could provide good ideas to process the data. Thanks!

Hi Shawn,

You seem to have a preferred orientation there. So, in one direction you have high-res, and the dominance of this view stretches your density.

Balance the views, induce them on cryo or add tilted data and you will fix it.



Hi JP,

Thank you very much for your suggestions. When I got this map, the first idea is that there must be a preferred orientation. However, we found that the test dataset collected from a 200 kV microscope with the same grads was able to be processed smoothly with no orientation issue, can not figure out what happened when we transfer to collect high-resolution data, maybe we lost important views during collection. I have tried to use the rebalance 2D class in cryoSPARC but didn’t work. I will try to figure out how to balance the views during sample preparation and data collection.
Thank you again!



Pretty challenging project of yours, I think…

If of interest, there is a script by Chris Russo called cryoEF that will give suggestions regarding the best tilt angles to fill the missing orientations.
Moreover, if your 90 kDa complex has an elongated form, it is possible that several of the views are just too thin to be properly picked and aligned… if that’s true, then you’d need to make it bigger somehow.
You can also review the process of selection - how did you get to that 150 k ptcls set - to see if the missing views were excluded at some point.
And from that same 150 k ptcls set, what happens if you run 3D classification asking for 4 or 5 classes? (or more if you wish) - the question here is how flexible the complex is.

I disagree that the main issue here is an underrepresentation of 2D projections for some angles. We had various datasets where the angular distribution looked similar to what you show, for particles maybe a bit bigger, and this did not prevent to reach an acceptable density map. When you have a 180° coverage that should work.

I have a few questions that might help you:

  1. How many particles you have after cleaning up your 2D classes ?
  2. How does your ab-initio look like ? Do you have clear secondary structure there ?
  3. What parameters you use for your ab-initio ?

If you have a lot of particles, you need maybe to clean it more. There are various discussions about parameters used for ab-initio that may came at hand for small particles. Have a look at those threads. There are also a couple of threads about parameters for NU-refinement, but you should think about this only when your ab-initio is satisfactory. Cheers

PS: I saw only after you had 150k particles. Conclusion: you need more (high-quality) particles. Collect more data and merge datasets ! Good luck

1 Like

Hi Carlos,

Thank you very much for your suggestions. We are considering collecting tilted data, but it may not solve the problem with such a small protein.
I have tried to do several rounds of 3D classification with 1-6 classes and 3-6 A resolutions but made no difference, it seems that there is no larger conformation change of the complex.
I am thinking about maybe I can focus on the process of particle selection for this data because I found there is one dominated view of 2D, maybe I need to throw away part of them to balance the views?

Best regards,

Hi Marino,

Thank you for your good suggestions! After 2D classes, I got about 400k particles, after 3D classes and heterogeneous refinement the number of particles goes to 150-200k, and I could get a good model after ab-initio and the model be much better after heterogeneous refinement. I believed the biggest issue for this data is the 3D refinement not working, I will definitely try more parameters and methods to do the 3D refinement.
Thank you!


I wonder if masking the RNA and protein separately and using local refinement for each might help?

Hi gdodge,

Thank you for your suggestion. I will try that, but for this protein-RNA complex may not work, since the RNA was most buried into the protein.


Hi Shawn,

I only mentionned CryoEF because I’m facing a similar problem, and you are right, from what people say, tilted data adds some complications, and it is not 100% sure to work in both our cases.

Thank you for bringing this up to discussion, I’ll keep following the topic.

All the best,