Difference between revisions of "Tomography processing notes"

From Helpful
Jump to: navigation, search
m (In general)
m (In general)
Line 45: Line 45:
 
* roughly known angles for each such image
 
* roughly known angles for each such image
 
: in IMOD: MRC header, text file, or given interval
 
: in IMOD: MRC header, text file, or given interval
* tilt axis angle
+
* tilt axis angle (=basically the rotation between sensor/image and tilt axis, so the direction it rotates on within the image)
: theoretically ideally perpendicular to sensor('s longest dimension), but rarely true for practical reasons
+
 
: if wrong, the coarse alignment will fail
 
: if wrong, the coarse alignment will fail
 
: ±5 degrees is fine because it's usually a parameter it's finding in fine alignment
 
: ±5 degrees is fine because it's usually a parameter it's finding in fine alignment
 
: ...but coarse alignment will probably fail if it's more than ~5 degrees off (because it's trying to accounting for the direction things squeeze in{{verify}})
 
: ...but coarse alignment will probably fail if it's more than ~5 degrees off (because it's trying to accounting for the direction things squeeze in{{verify}})
 
: IMOD: MRC header, or given value
 
: IMOD: MRC header, or given value
 +
<!--: theoretical ideal is perpendicular to sensor('s longest dimension), but rarely true for practical reasons-->
  
  

Revision as of 15:50, 10 January 2019

This article/section is a stub — probably a pile of half-sorted notes, is not well-checked so may have incorrect bits. (Feel free to ignore, fix, or tell me)

See also EM software notes

On gold fiducials

Apparently in theory there's

  • Au11 (0.8mn) a.k.a. undecagold
  • Au25 (0.9mn)
  • Au38 (1.1nm)
  • Au68 (1.3nm)
  • 1.4nm, often referred to as nanogold
  • Au102 (1.5nm)
  • 1.8nm (a.k.a. Ni-NTA–Nanogold(verify))
  • Au114 (2.0nm)
  • Much larger is colloidal gold (so these sizes are a little more approximate), often one of
    • 5nm
    • 10nm
    • 15nm
    • 20nm
    • 25nm

Apparently most commonly it's 5nm or 10nm, sometimes nanogold(verify).

See also "[Site-specific biomolecule labeling with gold clusters." (PMC3568671)



In general

The overall steps in reconstruction


A typical approach is described e.g. in imod. See e.g.

the overall diagram in "Automated tilt series alignment and tomographic reconstruction in IMOD"
See also "Tomography Guide for IMOD Version 4.9"
and a list of parts (some optional) in the batchruntomo man page.
and its tutorial https://bio3d.colorado.edu/imod/doc/batchExample.html


Required input

  • a series of images - a tilt series
  • roughly known angles for each such image
in IMOD: MRC header, text file, or given interval
  • tilt axis angle (=basically the rotation between sensor/image and tilt axis, so the direction it rotates on within the image)
if wrong, the coarse alignment will fail
±5 degrees is fine because it's usually a parameter it's finding in fine alignment
...but coarse alignment will probably fail if it's more than ~5 degrees off (because it's trying to accounting for the direction things squeeze in(verify))
IMOD: MRC header, or given value


Useful input:

  • knowing fiducial bead size eases finding them
  • Target bead amount helps using what you have.
  • Sample thickness
can be calculated from bead set (but not under all conditions, so a sensible fallback value can helps)
used to not make the back-projected volume much larger than the sample (roughly "you can stop this far away from the beads")
also relevant to rotating the sample within that volume to align with one axis, for more sensible slicing, and smaller volume



Steps

  • optional: dark/blank images
  • remove X-rays, pixel defects
useful for normalization step
IMOD: ccderaser
  • Rough alignment
basically, minimize it jumping about
otherwise not solving anything, but it assists fine alinment
(e.g. somewhat specialized cross-correlation)
IMOD: tiltxcorr, creates .preali
  • Finer alignment, by one of:
    • fiducial alignment (fiducials are small high-contrast things, typically gold beads)
    • fiducialless alignment, e.g. patch tracking / feature tracking
    • skipping (effectively assumes rough alignment is enough, which it basically never is)
  • combine aligned images into stack
  • correct CTF (optional)
only really required for higher resolution, which you may not need
  • remove fiducial markers from images
would be disturbing influence in tomogram
  • filtering
  • make tomogram from the stack
    • IMOD: weighted backprojection (WBP), or SIRT (Simultaneous Iterative Reconstruction Technique)
  • cut out things that aren't specimen (optional, but rather practical)
  • noise reduction via NAD (nonlinear anisotropic diffusion) filter (optional)
  • where applicable: combine dual axis tomogram
  • Optional: Subvolume averaging (if you have multiple of the same thing within a tomogram that you want to combine (sort of single particle style)


Notes:

  • Fiducial tracking can be done by manually telling it which correspond, or doing it via IMOD, or RAPTOR, or other.
Doing fine-detail alignment with fiducials is a nontrivial optimization problem.
Using more makes for finer detail in the alignment (averages and noise),
but also requires and kicking out the ones that seem to disagree with the consensus,
because a few bad cases would throw off otherwise detailed averages.
it may be easier to find/verify beads in 3D, but it's easier to remove beads in 2D, so you basically go through things twice.
IMOD: imodfindbeads, autofidseed, others

(semi)automated

Etomo runs you through the steps with some interaction in each.


batchruntomo is essentially a job manager

you give it directive files (adocs)
...with pre-set steps; you can run parts at a time (see -start, -end), useful for inspection before you continue, and to redo things with different parameters.

The list of directives: https://bio3d.colorado.edu/imod/doc/directives.html

Directives are sorted into:

  • setupset - mostly describes collection and dataset
  • com-param - parameters for programs (like in .com files)
  • run-time - parameters interpreted by batchruntomo or etomo (verify)

Unsorted

PEET (Particle Estimation for Electron Tomography) [1]


RAPTOR (Robust Alignment and Projection Estimation for Tomographic Reconstruction) used to be separate, and is now integrated, and a more automatic alternative. [2]