When curating particles for training a neural net picker (either Topaz or deep picker), it is very important that as many of the picks as possible are true positives, and hence curation of particle picks is important
However, it is often hard to see particles clearly as displayed in manual picker or inspect particle picks - or rather, it is much easier to see them clearly (and judge their centering) in the lowpass filtered, boxed and extracted particles shown in the Manual Picker log.
Would it be possible to either augment the existing inspect particle picks job or create a new job for viewing and curating individual particles?
Something almost identical to the existing select2D interface would work, I think - just allowing curation of individual lowpass filtered particles, rather than 2D classes. This would be very helpful for creating training data for neural net pickers, I think.
Cheers
Oli
I also wish such a tool to check individual particle after 2D classification.
I can see some representative thumbnail in 2D classification job -> Overview -> Output. But I wish I could check every particle.
Just bumping this and also showing what I mean.
You can see in the attached an example mic, and lowpass filtered images of extracted particles. For myself, I find it much easier to recognize well centered and non-aggregated particles in the boxed and extracted particles, as opposed to on the micrograph. This is particularly the case when the micrograph is crowded.
Some way to curate/prune extracted particles (and perhaps sort them by derived parameters, e.g. some kind of measure of variance or contrast, as well as class probabilities, shifts etc) would be useful when selecting particles for training picking models.
Cheers
Oli
In 4.1, 3DFlex integrates an option to select particles by scale factor. It might be useful to have this as a standalone particle curation utility, with the possibility of also selecting/sorting particles by other derived parameters (e.g. rotations/shifts after local refinement; variance/contrast; intensity prior to normalization; number of symmetry related copies remaining for this particle; etc).
1 Like
This would still be useful to have!
There is a way though to inspect individual particles, though it is a bit of a hack:
- Restack particles with a batch size of 1
- Import micrographs, selecting the restack directory of the job from (1) - treating the particles as micrographs, and setting output constant CTF.
- Run Manual Picker or Curate Micrographs using the output of (2).
e.g.:
This allows one to view individual particles, but it doesn’t really let one do anything with them, as all the metadata and CTF info is gone. But it is at least handy for simple inspection (and probably one could import CTF info somehow using CS-tools?). Particularly useful for inspection after signal subtraction, in my experience.
If it were possible to apply the same lowpass to the thumbnails, and sort by different metadata parameters (CTF fit, class probability etc), this could be quite useful:
Cheers
Oli
4 Likes
Thanks @olibclarke for posting the feature suggestion, which we recorded, and pick inspection workflow.
2 Likes
