Experiment with stitching for spatial omics datasets

[LucaMarconato]

Stitching is not supported in NGFF nor spatialdata, but in spatialdata we build on libraries like spatial-image (xarray) for which stitching is supported via https://github.com/multiview-stitcher/multiview-stitcher from @m-albert.

This track aims at experimenting with stitching in spatialdata.

1. (easy to prototype) we show how we can build an element that represents the stitched way doing some processing on the fly. “You define an element via stitching”. For company-use, with hundreds of tiles, this would be preferred (at the performance cost that on-the-fly stitching is going to be slower than loading a pre-saved dataset)
2. (more involved but close) Stitching could be done via transforming multiple elements to a single element. First there is an alignment step and then there is a summarized element. “You reduce the stitched elements into one”. May be preferred when one has a small amount of data and can afford to save the stitched version.

Notes

1. One can investigate the impact of doing multiprocessing with ray or joblib vs (or combined with) dask. This can be checked both for small and large arrays.
2. @m-albert could you suggest a dataset already used in https://github.com/multiview-stitcher/multiview-stitcher? For a spatial omics dataset we could use one of the datasets used in https://github.com/LieberInstitute/visiumStitched

[m-albert]

Thanks for this @LucaMarconato !

@m-albert could you suggest a dataset already used in https://github.com/multiview-stitcher/multiview-stitcher?

A relevant one I can currently think of would be the ashlar example dataset, processed here in multiview-stitcher. Will add more.

For a spatial omics dataset we could use one of the datasets used in https://github.com/LieberInstitute/visiumStitched

That's a great resource! Will be useful for experimenting with e.g. an example notebook.