Group by micrograph?

Tips
1

Hi,

Heterogeneous refinement works really well for removing junk, but it does so on a per particle level.

I’d like a way to identify whole bad (or worse than average) micrographs, reducing tedious and subjective manual screening, and I’m wondering whether performing a heterogeneous refinement with particles grouped by micrograph would do the trick - that is, instead of allowing for individual particles to swap between good and junk classes, to allow all particles belonging to a given micrograph to do so. I’m guessing this would only work (if at all) once the dataset is relatively clean and most obvious junk is removed.

Cheers
Oli

2

Hi @olibclarke :slight_smile: This is an interesting idea, but as you say it sounds like it would only start to become effective if the models (both good classes and junk classes) are already reasonable so that classification of all particles in a micrograph will give useful indication about the whole micrograph being junk.
When a micrograph is junk, does this usually manifest in sort of “low-quality” particles (I’m imagining cases like extreme ice thickness or motion blur) or is it usually because of some other characteristic like gold particles in the ice or bubbles or aggregation?

3

@apunjani it can vary, but I’m thinking mostly about motion blur and ice thickness - but really anything that systematically makes a micrograph worse than the bulk. I ask also because manual micrograph screening is tedious, subjective but currently still necessary in most cases, so if cryosparc could be used to identify the optimal set of micrographs that would be really useful.
Cheers
Oli

1 Like
4