The problem. Single-cell analysis has Scanpy: a mature, composable toolkit everyone shares. Spatial analysis, for a while, was a pile of one-off scripts. What you want is the same thing — a common object and a standard set of spatially aware operations — so analyses are reproducible and comparable across labs.

The idea. Squidpy builds on the AnnData world and adds the spatial layer: a neighbors graph based on physical coordinates, and spatial statistics on top of it — neighborhood enrichment, co-occurrence, Moran’s I for spatial autocorrelation, ligand–receptor analysis. It also carries an image container so histology and expression live together. The design goal is scalability and interoperability, not a new silo.

Why it matters. For a facility, the tool everyone already knows is the reproducibility win — shared vocabulary, shared object, less bespoke glue. Squidpy is where a lot of the earlier reading actually gets executed: the neighborhood graph here is the same spatial-context idea that BANKSY and the deconvolution methods each exploit differently. It’s the practical hub of the spatial stack.

Verdict. The sensible default and the one I’d build a facility workflow around. Its ceiling is that it’s a framework, not an answer — it gives you the operations, you still bring the biological questions and the QC judgment. Read it as the toolkit, then pair with SpatialData for the storage layer.