Appropriate parameters for homogeneous reconstruction #159
olibclarke
started this conversation in
Show and tell
Replies: 2 comments 10 replies
-
Thanks for doing this comparison and sharing the results! I was just thinking about testing your EMPIAR entry myself. Your commands look right -- I suspect that the default parameters to |
Beta Was this translation helpful? Give feedback.
9 replies
-
The box size should be parsed automatically by |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi,
I'm just getting started with CryoDRGN, so apologies if this is basic or if I am doing something stupid. To start with I figured I would try cryoDRGN ab initio homogeneous reconstruction, on a dataset I am familiar with (https://www.ebi.ac.uk/empiar/EMPIAR-11043/).
I first downsampled the particles from the consensus reconstruction like so:
cryodrgn downsample cryosparc_P1_J3493_013_particles.cs -D 128 -o ank_downsample_128.mrcs
Then, parsed CTF:
cryodrgn parse_ctf_csparc cryosparc_P1_J3493_013_particles.cs -o ank_ctf.pkl
Created a random subset of 20k images:
cryodrgn_utils select_random 710437 -n 20000 -o ank_downsample_20k.pkl
Then, ran ab initio reconstruction:
cryodrgn abinit_homo ank_downsample_128.mrcs --ind ank_downsample_20k.pkl --ctf ank_ctf.pkl -o ./abinit_homo_test >> abinit_homo_test.log
Resulting reconstruction:
https://www.dropbox.com/s/cfi6fusr5rvc9qq/cryodrgn_reconstruct.mrc?dl=0
As a comparison, I ran single class ab initio reconstruction in cryoSPARC, using default parameters, on a 20k random subset, also downsampled to 128px:
https://www.dropbox.com/s/bv3no0mbwe4uh1s/cryosparc_reconstruct.mrc?dl=0
For comparison, here is a direct reconstruction from the 20k subset, using the poses from the consensus refinement:
https://www.dropbox.com/s/zj6txx9zaa9j6f0/cryosparc_reconstruct_consensus_poses.mrc?dl=0
The results from cryosparc look a lot more detailed, and a lot closer to the consensus refinement. This makes me think I am doing something wrong. Is there something I should alter, in terms of either preprocessing or parameters for the ab initio reconstruction? Many thanks for any help or advice!
Beta Was this translation helpful? Give feedback.
All reactions