v0.1.0 — first public release

First public release of spatial-smooth — composable smoothing of gene-set signatures over space and cell state for single-cell and spatial transcriptomics.

Install

pip install "spatial-smooth[all]"

Only numpy, scipy, pandas and anndata are required; every backend is an optional extra, imported lazily (dm, embedding, plot, squidpy, kde).

What's in it

• One-line smoothing of a gene-set signature over physical space (KnnGaussian, the default), cell state (KompotGP over a diffusion map), or both composed — ss.smooth(adata, genes, name, steps=...).
• Composable pipelines: each step smooths the expression matrix over one embedding and hands the result to the next; shorthands ("spatial", "dm", "dm+spatial", "spatial-kde", "spatial-gp") or explicit Step objects.
• Smoothers: KnnGaussian (Gaussian kernel over k spatial neighbours), Kde (FFT Nadaraya–Watson on a fine grid), KompotGP (Gaussian-process regression).
• Compute once, plot forever: results are written into the AnnData (obs[name], obs[f"{name}_raw"], provenance in uns); plotting reads only those keys. ss.pl.signature / ss.pl.compare wrap scanpy and squidpy.
• Scale-invariant bandwidths (multiples of the median nearest-neighbour distance) and provenance that records the effective bandwidth, retained kernel mass, and a truncation warning when it bites.

Important

This package smooths for visualization. Smoothed values are spatially autocorrelated by construction — never run statistics (differential expression, clustering, correlations, p-values) on them. Every call also writes the unsmoothed score to adata.obs[f"{name}_raw"]; run statistics on those.

Docs

Documentation: https://settylab.github.io/spatial-smooth/ · Tutorial notebook and Concepts pages included.

Full API and tutorial: https://settylab.github.io/spatial-smooth/