search

bpowell122 / tomodrgn

Neural networks to analyze structural heterogeneity in cryo-electron sub-tomograms
GNU General Public License v3.0
18 stars 0 forks source link

Input from RELION-4 and from 3D subtomograms #2

Open rdrighetto opened 1 year ago

rdrighetto commented 1 year ago

Hi,

I am interested in trying tomoDRGN but I don't have Warp-generated stacks of tilts per subtomo for this dataset. So I have two questions:

  1. Would it be possible to provide as input RELION-4 metadata as defined in the tomograms.star and particles.star files (i.e. an optimisation set)?
  2. Alternatively, would it be possible to provide tomodRGN directly with 3D subtomograms (i.e. not individual tilts per subtomo)?

Thank you!

Best wishes, Ricardo

EuanPyle commented 1 year ago

Seconded, this would be very useful :)

procyontao commented 11 months ago

Agreed. I am interested to know this.

stefanoliver commented 5 months ago

Also expressing interest in this.

EuanPyle commented 5 months ago

Hi all, the latest version of RELION (v5) actually uses 2D stacks of images, so should be compatible with the cryoDRGN-ET, with a little tinkering. I am thinking of writing a wrapper so the two are compatible but I need to clarify one thing:

In RELION v1.4 (which WARP/M uses), does anyone know where the rlnDefocusU/V values, which are read by cryoDRGN per-particle, are derived from? Are they simply the DefocusU/V values from the central tilt image of the tilt series stack during CTFFIND? Or are these values adjusted per-particle based on Z-height? Or something else entirely?

logicvay2010 commented 3 months ago

Hi all, the latest version of RELION (v5) actually uses 2D stacks of images, so should be compatible with the cryoDRGN-ET, with a little tinkering. I am thinking of writing a wrapper so the two are compatible but I need to clarify one thing:

In RELION v1.4 (which WARP/M uses), does anyone know where the rlnDefocusU/V values, which are read by cryoDRGN per-particle, are derived from? Are they simply the DefocusU/V values from the central tilt image of the tilt series stack during CTFFIND? Or are these values adjusted per-particle based on Z-height? Or something else entirely?

In early version of RELION, the rlnDefocusU/V seems to be calculated per particle based on <defocus at Z=0 +/- a shift (defined by their geometric Z location) >. I am not sure whether they will refine it later in 3D refinement, probably not.