Sub-tomogram manipulation during extraction

[Guangyang Cai]

Hi all,

I was wondering if anyone had suggestions for the following questions:

1. For the original subtomo processing pipeline in relion published in https://www.nature.com/articles/nprot.2016.124, Tanmay offers a 2D classification workflow using Z-projected subtomos. Does Warp have an option to Z-project subtomograms during export? I know it has an "export images" option, but I understand that to export individual back-projected tilt-images rather than a projection through the subtomo down the Z-axis.

2. Is there a functionality out there to pre-multiply each subtomogram by its corresponding 3D CTF model? I'm trying to use 2D classification as an initial computational step in determining initial pose alignments for my subtomograms, and my subtomos contain mostly high-resolution features with little low resolution data (i.e. like amyloid fibrils). I would like to recover more of the high-resolution data in my subtomograms before Z-projecting them so that I can take advantage of the 2D classification pose assignments and map them back to the 3D subtomogram poses.

3. With the "Pre-rotate particles" option during subtomogram export, is there a way to track how each subtomogram has been transformed relative to its initial extracted coordinate system in the star file? It currently resets all of the angular and offset values to 0 after manipulating the subtomo, but it would be helpful to keep track of how things map back to the original coordinate system.

Thanks all!

-Andy

[teg...@gmail.com]

Hi Andy,

1. This workflow isn't implemented, and I can't think of a case where it might be superior to using full sub-tomos. If you really want to try it, running relion_image_handler with the --average option on every sub-tomo (possibly after changing the extension to .mrcs) should give you projections along Z. You could then get the CTF parameters by running sub-tomo export in Warp with the 'image series' option and number of tilts limited to 1, and modify the STAR file later to point to the projections. In Relion, you would need to specify that the CTF of the data is phase-flipped because the flipping is done during sub-tomo reconstruction in Warp. You could also run 3D classification with full sub-tomos using many classes and limiting the angular searches to Psi only (look into relion_refine command line options) to get somewhat similar behavior without going through projections.

2. Not sure how this would give you more high-resolution information. I don't see an obvious way to do this through Relion, so you'd have to find another way (e.g. Python & numpy) to FFT the sub-tomo volumes, multiply by the CTF, and IFFT back. If you don't square the CTF afterwards as well, amplitude correction will be messed up later.

3. Not implemented right now, though I agree this would be useful. The particle order in the STAR file stays the same, so you could just get the angles from one of the previous files.

Cheers,

Dimitry