A poorly refined result with surrounding "faces"

[AsDeadAsADodo]

Hi community.
I’m running into a recurring issue with my cryoSPARC refinement and would appreciate any advice.

The problem:
Despite dropping particle count to ~80k and multiple rounds of 2D cleaning, I still get extra surrounding faces/density in refined maps. The core seems okay in ab-initio, but refinement degrades it. Both of them have good viewing direction distribution.

2D classification class averages:

Ab-initio result:

Homo refinement:

What I’ve tried:

• Multiple rounds of 2D classification to remove junk

• Reduced particle stack to ~80k

Any suggestions would be greatly appreciated!

[CryoEM2]

your 2D suggest extreme compositional heterogeneity - many different protein forms. these failed refinements are the expected result, and though the euler distribution looks great in theory, in practice it suggests failed alignment and you actually want 4-6 distinct patches to cover the angles and as evidence that it could well-assign repeating views to a common stucture. I would suggest a large dataset, ab initio with many models, an improved picking strategy, and maybe a more pure sample in the first place.