search

ml-struct-bio / cryodrgn

Neural networks for cryo-EM reconstruction
http://cryodrgn.cs.princeton.edu
GNU General Public License v3.0
316 stars 76 forks source link

Wrong pixel size in landscape clustering output #142

Open priiteek opened 2 years ago

priiteek commented 2 years ago

The mean reconstructions of clusters from landscape analysis all have a pixel size of 1 Å/px even though a different value was given as input with cryodrgn analyze_landscape --Apix. All mrc-files in the "clusteringL2..." folder are at 1 Å/px. The same goes for the ones in "vol_pcs". The volumes in "kmeans1000" are correct though. I'm using cryoDRGN v1.1.0.

A quick fix is to change the pixel size in the header with relion_image_handler --i <input.mrc> --o <output.mrc> --force_header_angpix <correct_pixel_size>

But it would be great if the correct pixel size would be written to the header automatically.

Another thing that can also create some confusion is that the default box size of the reconstructions is 128 and the --Apix value applies to the reconstructions, not the input. So if training was performed on a stack with box=256 and a pixel size of 0.9 Å/px, the correct --Apix for cryodrgn analyze_landscape would be 1.8 Å/px.

It would make things much more convenient if the correct pixel size would migrate automatically instead of becoming 1.

zhonge commented 2 years ago

Thank you for reporting! @vineetbansal can you take a look?