Technology-specific advice

Sopa is designed to run on all spatial technologies at single-cell resolution, but some choices or parameters may be more or less suited to certain technologies.

Note

All this advice is for the API usage, but it also applies to the CLI and the Snakemake pipeline (update the usage accordingly or look at our per-technology Snakemake configs).

Xenium or MERSCOPE

For Xenium or MERSCOPE data, we recommend using a transcript-based segmentation tool, e.g., Baysor or Proseg. With the Xenium 5k panel, Proseg is recommended due to its fast results and high quality.

To use the prior 10X or Vizgen segmentation, you need to precise it in sopa.make_transcript_patches - with sopa>=2.0.4, use prior_shapes_key="auto" to automatically detect the prior and use it for Baysor/Proseg.

sopa.make_transcript_patches(sdata, prior_shapes_key="auto") # for proseg, also add patch_width=None

If using Baysor, you can also use area=20 to filter cells with an area below 20 um².

To filter low-quality cells, you can provide min_transcripts=20 to the aggregation.

CosMX

For CosMX data, the same advice as above is applicable, although you may experience some issues when reading the data with sopa.io.cosmx due to some frequent changes in the AtoMX exports. If so, please open an issue to improve this.

Visium HD

See here the full Visium HD specific tutorial.

MACSima or PhenoCycler

For spatial proteomics, Cellpose is needed. To avoid having artifacts outside of the tissue, we recommend running sopa.segmentation.tissue before running Cellpose. This way, Cellpose will run only inside the tissue.

For MACSima, we recommend to use diameter=35 pixels and min_area=400 pixels² to sopa.segmentation.cellpose. For PhenoCycler data, it will depend on the image resolution, so you'll need to choose a diameter (in pixels) that is biologically relevant.

During aggregation, you can for instance use expand_radius_ratio=0.1 to expand the cells, min_intensity_ratio=0.1 to filter cells with a too low intensity.

H&E or WSI

If you have only a H&E/WSI slide without spatial omics, you can also use Sopa, although many operations are dedicated to spatial omics. For instance, sopa.segmentation.tissue for tissue segmentation, sopa.segmentation.stardist for cell segmentation, sopa.patches.compute_embeddings and sopa.patches.cluster_embeddings for patches embeddings/clusters.

Though, if you have H&E with spatial omics, e.g. Xenium + H&E, in that case, we recommend segmenting the data on the spatial transcriptomics and aligning the H&E to performed joined operations as in this tutorial.

Sopa WSI extra

There is a Sopa extra for WSI, you can install it via pip install 'sopa[wsi]'

Other

For other technologies not listed here, please open a new GitHub issue.