Isonet for non-EM data

[Jan Groen]

Hello community

I have cryo-soft X-ray tomograms (voxelsize 100 A / 10 nm) and I would like to run them through IsoNet.

First picture below is just a small screenshot of the type of data. I am making a model (IMOD) to exclude all the dark lipid droplets. But then if I create the mask it looks like the second picture, I just have these lines, it does not find anything in term of membranes although the mitochondria are quite clear by eye. Curiously, if I do not give it a boundary model file it does pick up structures (lipid droplets). I also tried "lying" to isonet, saying voxelsize is actually 10 A but it results in the same mask.

I am just starting with IsoNet so I am not super experienced just yet. Could anyone with more experience give some input on how they would target this type of data? Testing all the parameters would take too long so I am hoping to get a little info before.

I am using --density_percentage 50 --std_percentage 50 and for refinment --iterations 30 --noise_start_iter 6,11,16,21 --noise_level 0.05,0.1,0.15,0.2

Thanks in advance

Jan Groen

[Jan Groen]

One extra thing came to mind now while writing, the contrast values for the X-ray data are not as spread as in EM. I am working with values between 0 and 1, for the extremes, the differences between membranes and background could be in the range 0.001 or smaller. I am not sure in what range IsoNet operates.

So I have a second question, I could normalize the range to be comparable to EM data but can I then use that training to correct the "original" datasets?

Thanks

Jan

[YUNTAO LIU]

Hi Jan,

I do not have experience processing soft X-ray data. I believe that the masking does not need to be perfect in IsoNet. Especially for cellular data, masking may not improve the performance of IsoNet. It may not be a problem in picking up the dark droplet. Keeping the original parameters for masking or even having a larger percentage for both parameters should be fine.

For IsoNet, you do not need to normalize the tomograms. It will normalize automatically. It will make about 90% of pixel values to be in the range between 0 and 1 during normalization.  

--
You received this message because you are subscribed to the Google Groups "IsoNet" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isonet+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/isonet/379d57e4-ebfc-4d8f-9cf0-41c49070dcf2n%40googlegroups.com.

--

Best Regards,

Yuntao Liu,  Postdoc.

California NanoSystem Institute

University of California Los Angeles