pystatial 0.1.1

A pure-Python re-implementation of Bioconductor Statial for spatial cell state analysis

py-statial

A pure-Python re-implementation of Bioconductor Statial for spatial cell state analysis.

Install

pip install pystatial

Quickstart

import statial
import pandas as pd

# Load spatial data
cells = pd.read_csv("cell_metadata.csv")

# Calculate distances between cell types
import anndata as ad
adata = ad.AnnData(obs=cells)
adata = statial.get_distances(adata, max_dist=200)

# Calculate abundances
adata = statial.get_abundances(adata, r=200)

# Run Kontextual analysis
result = statial.Kontextual(
    cells=cells,
    r=50,
    from_types="Macrophages",
    to_types="Keratin_Tumour",
    parent=["Macrophages", "CD4_Cell"],
    image=["6"],
)

Function mapping (Python ⇄ R)

Python function	R function	Description
get_distances()	getDistances()	Pairwise cell type distances
get_abundances()	getAbundances()	Cell type abundances (K-function)
calc_contamination()	calcContamination()	RF-based contamination scores
Kontextual()	Kontextual()	Conditional spatial relationships
kontext_curve()	kontextCurve()	Kontextual over radii range
kontext_plot()	kontextPlot()	Plot Kontextual results
calc_state_changes()	calcStateChanges()	Linear model state changes
make_window()	makeWindow()	Create observation windows
parent_combinations()	parentCombinations()	Parent-child combinations
get_parent_phylo()	getParentPhylo()	Extract phylo tree structure
prep_matrix()	prepMatrix()	Convert results to matrix
get_marker_means()	getMarkerMeans()	Average marker expression
relabel()	relabel()	Permute cell labels
relabel_kontextual()	relabelKontextual()	Permutation testing
is_kontextual()	isKontextual()	Validate kontextual result

Benchmark

Dataset: Keren et al. 2018 MIBI-TOF breast cancer (data("kerenSCE"), patient 6) — 57,811 cells, 10 images, 17 cell types. Exported via colData() from R's SingleCellExperiment to CSV.

Environment: Python 3.9.13 + scipy 1.8 + sklearn 1.0 vs R 4.5.2 + Statial 1.11.6. All timings are 3-run means.

Parity — vs R reference

Function	Metric	Value	Threshold	Pass
get_distances	max abs error	1.33e-11	1e-8	✅
get_abundances	max abs error	0.0	1e-8	✅
Kontextual (L-function)	max abs error	7.11e-15	1e-8	✅
Kontextual (value)	relative error	5.14%	10%	✅

Core numerical functions match R at machine precision. The ~5% difference in Kontextual values comes from cKDTree vs spatstat closepairs spatial indexing.

Speed — Python vs R

Function	Python (s)	R (s)	Speedup
get_distances	0.84	7.48	8.9×
get_abundances	5.47	7.25	1.3×
Kontextual (img6)	0.09	0.20	2.2×
Total	6.39	14.93	2.3×

get_distances benefits most from cKDTree spatial indexing (8.9x). get_abundances is comparable (bottleneck is per-cell counting loop).

Citation

If you use Statial in your work, please cite:

Ameen, F., Iyengar, S., Qin, A., Ghazanfar, S., & Patrick, E. (2022). Statial: A package to identify changes in cell state relative to spatial associations.

License

GPL-3.0 (matching upstream R package)