Initial Vignettes

[kevinyamauchi]

This issue is to discuss the initial vignettes we want to make in order to demonstrate the utility and usage of SpatialData.

Here are the things I think we want to highlight:

• uniform data API makes it easy to use the same pipeline for different data sources
• transforms and coordinate systems allow data from multiple sources to be aligned and analyzed together
• interoperable data representation means that people can view and interact with their data on multiple platforms (e.g., web and desktop) without moving or transcoding.
• data can be interactively viewed and annotated

Blockers

Based on the vignettes below, these are the features that need to be implemented.

• [ ] aggregate by polygon (vignette 1) (luca)
• [ ] aggregate by label (vignette 1) (luca)
• [ ] Vitessce loading SpatialData (vignette 1)
• [ ] pytorch data loader from image (vignette 2) (waiting: after spatial query)
• [ ] bounding box spatial query (vignette 2) (kevin)
• [ ] static plotting (vignette 1, 2, 3) (tim)
• [ ] rasterization (unassigned)
• [ ] downscaling and downscaling a multiscale (unassigned)

Vignette 1: flexible pipelines operate on multiple data sources

data source: Xenium (luca), CoxMX (giovanni), Merscope (giovanni) features highlighted:

• uniform data API makes it easy to use the same pipeline for different data sources
• interoperable data representation means that people can view and interact with their data on multiple platforms (e.g., web and desktop) without moving or transcoding.
• data can be interactively viewed and annotated

Blockers

• aggregate by polygon
• aggregate by label
• Vitessce loading SpatialData
• static plotting

Analysis steps

1. Load the data
2. View data
3. Cluster cells
4. Interactively explore clusters in napari-spatialdata
5. annotate regions with polygons in napari-spatialdata (e.g., annotate an anotomical region)
6. aggregate based on the drawn regions
7. perform differential expression or compare cell type compositions in the annotated/aggregated regions
8. write results to disk
9. view results in Vitessce (in the notebook directly?)

Vignette 2: multi-FOV analysis

data source: [Erickson et al., Nature, 2022] (https://www.nature.com/articles/s41586-022-05023-2) (giovanni) features highlighted:

• transforms align data from different sources
• integration with the scientific python ecostem (specifically image analysis + transcritpomics)

Blockers

• pytorch data loader from image
• bounding box spatial query (vignette 2)

Analysis steps

1. Load data 1.1 Alignment with napari
2. View data
3. Cluster/classify visium spots
4. Extract images under visium spots into pytorch data loader
5. learn to predict spot cluster based on H&E image (not sure this will work well...might want to have a back up idea)
   7. code from squidpy
6. predict on the rest of the image

Vignette 3: multimodal integration

data source: Xenium + Visium features highlighted:

• transforms align data from different sources

Analysis steps

1. Load data
2. View data
3. Align
4. Do analysis

[kevinyamauchi]

Done!

https://spatialdata.scverse.org/en/latest/tutorials/notebooks/notebooks.html