The reconstruction using tomodrgn backproject_voxel is incorrect

[xinsheng44]

when I use tomodrgn backproject_voxel with the EMPIAR-10499, the output is just a blurry, spherical shape. I’ve tried multiple variations of the command, but none produce the correct result.

tomodrgn backproject_voxel \ input_v1/bin1_2d_tomodrgn_optimisation_set.star \ --output output/01_backproject_v1/backproject_70S_doseweight.mrc \ --datadir ../particleseries \ --uninvert-data \ --flip \ --recon-dose-weight \ --recon-tilt-weight

tomodrgn backproject_voxel \ input_v1/bin1_2d_tomodrgn_optimisation_set.star \ --output output/01_backproject_v1/backproject_70S_doseweight.mrc \ --datadir ../particleseries \ --flip \ --recon-dose-weight \ --recon-tilt-weight

tomodrgn backproject_voxel \ input_v1/bin1_2d_tomodrgn_optimisation_set.star \ --output output/01_backproject_v1/backproject_70S_doseweight.mrc \ --datadir ../particleseries \ --recon-dose-weight \ --recon-tilt-weight

The figures also show these incorrect results.

The particle extraction commands were

WarpTools ts_export_particles \ --settings warp_tiltseries.settings \ --input_directory relion/Refine3D/job004 \ --input_pattern "particle.star" \ --output_star relion/bin1_2d_tomodrgn.star \ --output_processing warp_tiltseries/particleseries_tomodrgn \ --coords_angpix 3.401 \ --output_angpix 1.7005 \ --box 190 \ --diameter 220 \ --relative_output_paths \ --2d \ --device_list 0 \ --perdevice 6 and WarpTools ts_export_particles \ --settings warp_tiltseries.settings \ --input_directory relion/Refine3D/job002 \ --input_pattern "particle.star" \ --output_star relion/bin1_3d_tomodrgn.star \ --output_processing warp_tiltseries/subtomo_tomodrgn \ --coords_angpix 3.401 \ --output_angpix 1.7005 \ --box 190 \ --diameter 220 \ --relative_output_paths \ --3d \ --device_list 3 \ --perdevice 6 respectively.

However, I used the following command (same particle dataset) to generate the desired result: mpirun -n NUM_MPI_PROCESSES relion_reconstruct_mpi --i particleseries.star --o reconstruct_relion.mrc --ctf As shown in the attached figures, one reconstruction is the raw result and the other is a smoothed version.

[zhonge]

Thank you for the detailed post! Please file an issue on the tomodrgn github repo. I believe they adapted our backproject_voxel script, but I can't speak to what changes they made.

The cryoDRGN repo supports heterogeneous STA (see our cryoDRGN-ET paper for more info). I would recommend trying out cryodrgn backproject_voxel with the --tilt option, and we're happy to help diagnose any problems you run into!

Here's an example usage from cryodrgn backproject_voxel -h:

# `--tilt` is required for subtomogram datasets; you can further control how many tilts
# are used per particle using `--ntilts`
# `--datadir` is generally required when using .star or .cs particle inputs

$ cryodrgn backproject_voxel particles_from_M.star --datadir subtilts/128/ \
                             --ctf ctf.pkl --poses pose.pkl \
                             -o backproject-tilt/ --lazy --tilt --ntilts 5

[xinsheng44]

Thank you for the detailed post! Please file an issue on the tomodrgn github repo. I believe they adapted our backproject_voxel script, but I can't speak to what changes they made.

The cryoDRGN repo supports heterogeneous STA (see our cryoDRGN-ET paper for more info). I would recommend trying out cryodrgn backproject_voxel with the --tilt option, and we're happy to help diagnose any problems you run into!

Here's an example usage from cryodrgn backproject_voxel -h:

# `--tilt` is required for subtomogram datasets; you can further control how many tilts
# are used per particle using `--ntilts`
# `--datadir` is generally required when using .star or .cs particle inputs

$ cryodrgn backproject_voxel particles_from_M.star --datadir subtilts/128/ \
                             --ctf ctf.pkl --poses pose.pkl \
                             -o backproject-tilt/ --lazy --tilt --ntilts 5

I'm really grateful for your help!

Initially, I planned to test cryoDRGN-ET, but unfortunately, it doesn't support CTF-multiplied images.

It only works with datasets from Warp/M (Windows) and doesn't support datasets generated by WarpTools.

What a pity!

#429

[zhonge]

Gotcha, we can certainly prioritize this feature. It might be super helpful to get some example files, and we can reach out to you if needed!