Artefacts in corrected tomograms (plastic sections)

[mvorlander]

Hi there,

I am trying to run (for collaborators) isonet on tomograms acquired from heavy metal stained plastic sections (so this isn't cryo data, no ctf correction was performed). The "corrected" tomograms display strong "pixelation" artefacts, see screenshots attached. Here are the commands I used to get there. I would be happy about suggestions what is going wrong here. More generally, has anyone tried IsoNet on non-cryo data?

isonet.py prepare_star tomoset --output_star tomoset.star --pixel_size 4.51

isonet.py extract tomoset.star

isonet.py extract prealigned.star

isonet.py refine subtomo.star --gpuID 0,1,2,3 --iterations 30

isonet.py predict tomoset.star results/model_iter30.h5 --gpuID 0

Thanks a lot,
Best,
Matthias

[procyontao]

Hi Matthias,

Sorry for late reply.

The pixel size in IsoNet is in Angstroms. I suggest you bin the tomogram to bigger pixel size e.g. >10A/pixel, or maybe 20A/pixel for plastic sections.

May I know what is the rough size for the each patch artifact? If it is about 64x64 pixels, that is caused by the small cropsize. I would try "isonet.py predict tomoset.star results/model_iter30.h5 --gpuID 0 --crop_size 128" to see whether this artifact still exists.

We tried IsoNet using the plastic sections tomograms in IMOD tutorial data, and it works well.

Best,
Yuntao]