Spatial count Operator

Description

Performs a spatial count using neighbourhood analysis. It counts the frequency of neighbour cluster per cell using scikit-learn's NearestNeighbors algorithm.

Usage

Input	.
y-axis	number, marker expression values (matrix)
column (1st)	text, image identifier (e.g., filename)
column (2nd)	number, unique cell identifier (e.g., Object)
column (3rd)	number, X centroid position (e.g., centroid-0)
column (4th)	number, Y centroid position (e.g., centroid-1)
colors	text, phenotypic cluster assignment
row	text, marker names (optional)

Settings	.
method	radius or knn, default radius
radius	number, setting for the size of radius to use, default 80
knn	integer, setting for the number of neighbors, default 10
n_clusters	integer, number of neighborhood clusters for k-means, default 8

Output	.
neighbourhood_cluster	integer, the neighborhood cluster assignment (0 to n_clusters-1)

Details

Performs spatial neighborhood analysis using the scimap library:

1. Reads marker expression data and cell metadata from Tercen projection
2. Uses column factors in order: imageid, cellid, x_coordinate, y_coordinate
3. Gets phenotype/cluster information from the colors projection
4. Counts neighbor phenotypes within a radius or using k-nearest neighbors
5. Performs k-means clustering on the spatial count results to identify neighborhood regions
6. Returns one neighborhood cluster assignment per cell

The operator uses:

• sm.tl.spatial_count() - counts phenotype frequencies in neighborhoods
• sm.tl.spatial_cluster() - clusters cells based on their neighborhood composition

Note: The operator uses generic column mapping based on factor order, not hardcoded column names. This makes it flexible for different datasets.

Example Output

Each cell gets one neighborhood cluster assignment:

.ci, .ri, neighbourhood_cluster
0, 0, 0
1, 0, 6
0, 1, 0
1, 1, 6
2, 0, 2

Cells with similar neighborhood compositions are assigned to the same cluster (0 to n_clusters-1).

The output maintains the crosstab structure with .ci and .ri indices for proper integration with Tercen workflows.