py-TranspaceR

Python port of TranspaceR — Statistical analysis of Spatial transcriptomic data.

Correlation Benchmark (Python vs R)

Function	Pearson r	Max Abs Error
C_normalisation	1.000000	4.10e-05
Otsu_thresholding	—	3.54e-05
colvars_sparse	1.000000	4.40e-05
Get_variogram_map	Deterministic match	0
Get_isotropic_vario	1.000000	0

All outputs are highly consistent with R references, with errors within floating-point precision.

Speed Benchmark (39,047 cells x 539 genes)

Function	R Time	Python Time	Speedup
C_normalisation	1.47s	0.157s	9.4x
Otsu_thresholding	0.31s	0.031s	10.0x
colvars_sparse	1.72s	0.011s	156x
Get_variogram_map	0.02s	0.0004s	50x
Get_isotropic_vario	0.01s	0.0004s	25x
Total	3.53s	0.20s	17.7x

Why faster

• NumPy/SciPy compiled C backend vs R interpreted execution
• Direct CSC sparse matrix memory layout access
• Broadcasting replaces R's row-wise apply loops

Installation

pip install py-TranspaceR

With optional dependencies:

pip install "py-TranspaceR[all]"

Quick Start

import transspacer as ts
import numpy as np
import pandas as pd

# Load data
expr = pd.read_csv("Expression_file.csv.gz", index_col=0)
meta = pd.read_csv("Meta_data.csv", index_col=0)

# Cell-size normalisation
normed = ts.c_normalisation(expr.values.astype(float), meta["Area"].values)

# Otsu thresholding
threshold = ts.otsu_thresholding(np.log10(expr.values.sum(axis=1) + 1))

# Variogram analysis
result = ts.compute_variogram(normed, meta["cell_centroid_x"].values,
                               meta["cell_centroid_y"].values)

# Geary's C spatial autocorrelation
gc = ts.geary_c_score(normed, coords, pvalue_threshold=0.01)

# Clustering
labels = ts.cell_clustering_function(pca_data, K=10, resolution=1.0)

Modules

Module	Description
fft_utils	fftshift, ifftshift, pad_definitor
normalization	C_normalisation cell-size normalisation
sparse_utils	Sparse matrix column variance, group aggregation
variogram	FFT variogram map, variogram model fitting
spatial_stats	Geary's C, NB excess variance / excess zero score
clustering	KNN + Leiden/Louvain clustering, UMAP
gene_selection	log2FC, gene set union
qc	Otsu thresholding, QC gene filtering
plotting	Spatial visualization, heatmaps, UMAP plots

Testing

pytest tests/ -q
# 29 passed

Dependencies

Core: numpy, scipy, scikit-learn, pandas, matplotlib

Optional: umap-learn (UMAP), python-igraph + leidenalg (Leiden clustering), statsmodels (FDR correction)

License

MIT