Spatial - Perform cell-type deconvolution analysis

[adkinsrs]

This was suggseted by Litao Tao at the HRP meeting after I presented on spatial transcriptomics with respect to gEAR

Brief tutorial we can use
https://squidpy.readthedocs.io/en/stable/notebooks/tutorials/tutorial_tangram.html

Currently we have planned the ability to project a novel set of single-cell clusters (via marker genes) into a reference spatial dataset to annotate the clusters. However, for spatial datasets and single-cell references it is not as straightforward. Generally with projection we are assigning the best projected annotation out of a series of probabilities, whereas as you move around the spatial region (basal to apical, for example) the region may be a proportion of cell types rather than a single one.

Further complicating things is that platforms that are bin-based, such as Visium may require image segmentation to determine how many cell nuclei are in each bin. Cell-type deconvolution would allow us to convert the bin unit into a proportion of cell-type assignments based on what we see in the single-cell reference.