[3dem] "Recenter" option in Relion during particle extraction

[Takanori Nakane]

Hi,

During refinement, particles are translated to be
consistent with the reference map.
These shifts are stored in the rlnOriginX/YAngst
columns.

When you recenter to (0, 0, 0), the center of particle boxes 
(rlnCoordinateX/Y) is moved such that
rlnOriginX/YAngst are as close as possible to zero.
(They cannot be exactly 0, because rlnCoordinateX/Y
are integers; sub-pixel shifts remain in
rlnOriginX/YAngst)

If you run relion_reconstruct on the re-extracted
particle STAR file, it will regenerate the same map
as the input run_data.star.

When you recenter to somewhere else (x, y, z) and run
relion_reconstruct, the point (x, y, z) of the
old map will move to the origin of the new map.
To do this, the center of particle boxes is
moved. For example, if a particle is in the top view
(XY plane on ice) and you re-center to (10, 0, 0),
the box will be moved horizontally (along X).
If you recenter to (0, 0, 10), the box will not move
at all, because Z direction does not matter in
projection. For general view angles, the box will move
by the projection of (x, y, z) onto the micrographs.

When you re-center to non origin, you HAVE TO SHIFT
the reference map with the same amount. Otherwise,
the following refinement will move particles back
to the old origin, or crashes by "no orientation
found" if the shift is too large beyond the
translation prior.

>   * If I use XYZ (which represent the difference between voxel center
>     and map’s center of mass) offsets for extraction and still use the
>     same starting reference then I get a perfectly centered
>     reconstruction, although it is very distorted and compressed
>     relative to what it should be,
>   * If I use XYZ offsets for extraction and also apply them to my
>     starting reference (for example: “relion_image_handler --i
>     reference.mrc --o reference_xyz.mrc --shift_x x --shift_y y
>     --shift_z z” or “relion_image_handler --i reference.mrc --o
>     reference_-xyz.mrc --shift_x -x --shift_y -y --shift_z -z”) then the
>     resulting reconstruction does not make any sense.

I don't understand what is happening here.

Just to make sure: The centering coordinate is in
pixels of the reference and the origin is at the
center of the box (box_size / 2 + 1). It has nothing
to do with the center of mass.

> Is there any clear answer preferably with an example about when the
> Recenter option is useful?

The most typical use is to re-extract particles
at a smaller pixel size after Class3D or Refine3D.
When doing this, re-center to origin (0, 0, 0).

The next case is to re-center particles
into the domain of interest. Suppose
your protein consists of two domains A and B,
and their relative orientations are flexible.
Even after classification, there remain some
flexibility. If you re-center to the the domain A
and put a mask around it, you will improve the
resolution of A, while sacrificing B. Next,
subtract A and re-center to B. Now you will have
better reconstruction for B.
(Multibody refinement internally performs something
similar to this but doing this manually sometimes
gives better results, although time consuming)

Best regards,

Takanori Nakane

On 2020/06/03 23:14, Iozzo, Yuri /US wrote:
> Hi everyone,
> 
> I wonder if someone can explain or can point me to a good description to 
> understand the “recenter” option when you extract the particles in 
> Relion. After years of use I am not sure what exactly it does.
> 
> Here is my experience with it:
> 
>   * If I do not use any offsets for particle extraction and for my
>     starting reference MRC file then I get a reconstruction usually
>     close to the coordinates of starting reference,
>   * If I use XYZ (which represent the difference between voxel center
>     and map’s center of mass) offsets for extraction and still use the
>     same starting reference then I get a perfectly centered
>     reconstruction, although it is very distorted and compressed
>     relative to what it should be,
>   * If I use XYZ offsets for extraction and also apply them to my
>     starting reference (for example: “relion_image_handler --i
>     reference.mrc --o reference_xyz.mrc --shift_x x --shift_y y
>     --shift_z z” or “relion_image_handler --i reference.mrc --o
>     reference_-xyz.mrc --shift_x -x --shift_y -y --shift_z -z”) then the
>     resulting reconstruction does not make any sense.
> 
> Is there any clear answer preferably with an example about when the 
> Recenter option is useful?
> 
> Thanks,
> 
> Yuri Iozzo
> 
> Sanofi
> 
> 
> _______________________________________________
> 3dem mailing list
> 3dem at ncmir.ucsd.edu
> https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem
>