questions on voxels and cropping spatial data about tangram HOT 1 CLOSED

Real thanks you for your interest giovp.

A Visium voxel is a circle with 50 um diameter which (typically) contains more than a cell. If you look at the Tangram paper, Fig 4a, grey circles defines voxel. In the voxel-by-gene data, the x and y coordinates found in the obs dataframe in the AnnData refer to the voxel centers (a.. "spot", if you want). Yet, the corresponding counts in the matrix X are obtained from all molecules within the voxel (hence, within a circle centered on the "spot" but with diameter 50 um).

You do not need any prior on cell density in order to map your data, BUT, if you wish to deconvolve (ie to obtain prediction for each single cell within a voxel) then you do.

Yes the prior on the cell density (or cell counts, you can provide either) has the shape (voxels,).

You can run out of memory with Visium data as well. If you have a lot of single cells, or you are mapping over a large portion of tissue, then you run out of RAM. If that so, we suggest to partition the dataset.

I am closing this issue, but if you think something is not clear please re-open and ask again.

Ciao,

from tangram.