COMET IMAGE ANALYSIS WITH SOPA

Setup and Installation

Assumes a GPU is available for the segmentation, and that the cpsam model fro cellprofiler4 is present at: ~/.cellpose/models/cpsam The cpsam model can be downloaded from: https://huggingface.co/mouseland/cellpose-sam/resolve/main/cpsam?download=true)

The following packages are used, see requirements.txt:

sopa
spatialdata_plot
spatialdata
matplotlib
scanpy
numpy

Tested on Python 3.11.2

Workflow

Preprocesing

1. Load the COMET image.
2. Set the chanenl names.
3. Segment the tissue on a lower resolution pyramid level, to identify the areas to perform cell segmentation on.
4. Divide the tissue in patches, to perform the segmentation on.
5. Plots: DAPI Image, tissue to be segmented, and segmentation patches to check the areas that will be segmented

Segmentation

This step needs to be performed on a GPU node. Assumes that the cpsam model fro cellprofiler4 is present at: ~/.cellpose/models/cpsam The cpsam model can be downloaded from: https://huggingface.co/mouseland/cellpose-sam/resolve/main/cpsam?download=true)

1. Cellpose segmentation with the cpsam model
2. Make ann AnnData object with the mean pixel value of each marker in each cell

Analysis

If the preprocessing and segmentation have been performed with the nf-core version of sopa, the SpatialData out folder in zarr format can be opened and the analysis started from here.

1. Produces an expression table withou DAPI for the UMAP plotting
2. Generate a NN neighbour grap and the UMAP embeddings
3. Assign the cells to specific celltypes using manual thresholds
4. Plots:
   • UMAPs
   • Violinplot
   • Dotplot
   • Heatmap

Output

Extract the bounding boxes of the 3 largest pieces of tissue:

1. Calculate the scaling factor between the pyramid level used for tissue segmentation and the full resolution
2. Get the bounding boxes from the regions_of_interest ROIs polygons we obtained from the tissue segmentation
3. Scale the bounding boxes to match the full resolution image
4. Calculate the area to find the 3 kargest bounding boxes
5. Use the bounding boxes to crop the spatialdata object
6. Write out zarrs for visualization with napari

Output and visualization

The output consists of separate spatialdat objects, one for each sample. The spatialdata objects contain:

• Image
• Cell plygons
• Annotated expression table

These objects can be visualized with napari using the napari-spatialdata plugin