Hello, we are trying to mix multiple dataset with different pixel sizes together. We don’t want to rescale the same pixel size. Instead, I was wondering if I could use a consensus size say 1.32 and the actual pixel size is indeed [1.317, 1.298, 1.322…], giving each dataset different unique_id and change the matrix in ‘ctf/anisomag’ [x1, x2, x3, x4]. From the RELION3.1 paper, I guess x2 and x3 represents streching angle and I could leave them as 0. I wonder how I should set x1 and x4, for example, ‘blob/psize_A’ for exp_group 1 is the consensus 1.32, and the resulting adjusted pixel size could be 1.317? Thanks a lot.

A following up question:

If we want to do this way, should we re-do the motioncorrction using the stated consensus pixel size 1.32, determine the ctf parameter based on this pixel size and then add the anisotropic matrix? Thanks a lot.

Hi @nym2834610,

Unfortunately, this workflow will not produce the expected behaviour and for the moment being, it is not possible to use anisotropic magnification coefficients to combine datasets of substantially (i.e. greater than a few percent) different pixel sizes. The reasons are due to the conventions by which we compute the CTF, which implicitly fitted to the values of the anisotropic magnification matrix such that arbitrary modification of it will render the defoci, spherical aberration, etc. incorrect. The use of anisotropic magnification to do this is on our roadmap but is a larger project than we initially anticipated due to this.

In the meantime, please see recent discussions on the forum for alternative options:

- Merging datasets - #2 by DanielAsarnow
- Merging two data sets with different pixel size - #2 by apunjani

Michael

okay, thanks a lot for letting me know.