Grid in MCore refine_imagewarp / refine_volumewarp

[B Wimmer]

Hi all, 

I am currently refining a test dataset of ribosomes from FIB lamellae, acquired on a K3. The tilt axis is ~ 90deg, and along the long axis of the micrograph. The dimensions of the raw micrographs are thus x = 6,000 px and y = 4,000 px. The reconstruction is rotated according to the convention that the tilt axis lies on Y, and thus has the unbinned dimensions 4k x 6k x 3k px. So far, so good. 

Now my question: When I refine imagewarp and volumewarp in these tomograms using MCore, I have to define the number of patches along x, y (and z, for volume warp). Does this relate to the original xy dimensions of the micrographs (ie. x is the longer axis), or the rotated dimensions of the reconstruction (ie. y is the longer axis)? 

Intuitively, I would have assumed that the latter is the case, meaning that the x and y patches in M relate to the final reconstruction. However, in the tutorial on the website, which seems to have a similar setup, the volume patches are given as "6x4" (https://warpem.github.io/user_guide/warptools/quick_start_warptools_tilt_series/#running-m-with-mcore). Is this a deliberate choice? How do people typically choose their patch sizes for M refinement? 

Thanks in advance for any advice and insights, 

Best,

Benedikt

[Warp Bot]

Hi Benedikt,

The two grids live in different coordinate systems:

1) Image warp (--refine_imagewarp XxY) is defined in the raw image reference frame, i.e. the original micrograph before any tilt-axis rotation. For your 6000x4000 px K3 images, 6x4 is the natural choice. This is why the tutorial also uses 6x4 -- it reflects the shape of the raw images, not the tomogram.

2) Volume warp (--refine_volumewarp XxYxZxT) is defined in the tomogram reference frame, where the tilt axis has been aligned with Y. For your data with a ~90 deg tilt axis, the 6000 px dimension of the original micrograph maps onto the Y axis of the tomogram, and the 4000 px dimension maps onto X. So if you want to match the aspect ratio of your reconstruction, you would use something like 2x3x2x10 (fewer patches along X, more along Y). The documentation for M describes this explicitly: "6000x4000 px images with a tilt axis angle close to 90 or 270 will need e.g. a 2x3 model."

In summary: image warp follows the raw micrograph shape (6x4 for your data), volume warp follows the tomogram shape (where X is the shorter axis for a ~90 deg tilt axis). The 6x4 in the tutorial is for image warp and is deliberate.

As a general approach, start conservatively (e.g. 1x1 image warp, no volume warp) and increase resolution iteratively. Optimal spatial resolution depends on the number of particles per tilt series and available signal per particle.

Best,

-- Warp Bot
THIS IS AN AUTOMATED MESSAGE GENERATED BY AN LLM. IT MAY OR MAY NOT SOLVE YOUR PROBLEM. IF YOU'D LIKE TO SPEAK TO A HUMAN, SAY SO IN YOUR MESSAGE.

--
You received this message because you are subscribed to the Google Groups "Warp" group.
To unsubscribe from this group and stop receiving emails from it, send an email to warp-em+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/warp-em/6bf1fb92-1cf2-419f-8203-71ebc17827e4n%40googlegroups.com.