Resolution worsens after non-uniform refinement

I am processing a apoferritin data collected on 300kV Krios (pixel size 0.5A). I did multiple rounds of 2d classification to remove junk.

Then did abinitio(3 class and 2 class). Ab initio didnt generate good model, so I imported a apoferritin model from EMDB and used it as one of the input volume in heterogenous refinement. This time I got a pretty good looking structure at 4.06A.

But when I did NU refinement of this structure, resolution decreased to 6.22A. I have used default settings in all the job type. I have total 75,570 particles.

What am I doing wrong?? Can someone suggest?? Do I have to play with settings or do more particle cleaning.

Your 2D classes looks great and I don’t think you need to spend more time removing junk right now. I would focus on making ab-inito work, with the particles you have, instead of importing 3D volumes as references.

The easy way to make ab-initio work is probably to enforce the O symmetry of Apoferritin. 2 classes from 10k particles is usually plenty.

If you want to make ab-initio work in C1, while the particle has high symmetry, you could try changing a few settings:

  • Change the “Maximum Resolution” from 12 Ă… to ~8 Ă….
  • Turn off “Center structures in real space”.
  • Turn off “Enforce non-negativity”.

Either way, once you have generated a good ab-initio volume to refine agains, you should use that O symmetry during your homogeneous (or NU) refinement.


Your 3D slices don’t look like they’re aligned, or that you’re imposing symmetry.

As (apo)ferritin is octahedral, enable O symmetry. :wink:

From 75,000 particles at 0.5Angpix, you should be seeing <2A (which is possible with ~3,000-4,000 particles).

1 Like