pycelladmix 0.1.0

Python port of cellAdmix — evaluating and correcting cell admixtures in imaging-based spatial transcriptomics data

py-cellAdmix

Python port of cellAdmix — evaluating and correcting cell admixtures in imaging-based spatial transcriptomics data.

Install

pip install pycelladmix

Quickstart

import celladmix
import pandas as pd

# Load your spatial transcriptomics data
df = pd.read_csv('molecules.csv')  # x, y, z, gene, cell, celltype, mol_id

# Run the pipeline
ca = celladmix.CellAdmix(molecule_df=df)
ca.run_knn_nmf(k=5, h=20)
ca.run_crf_all(num_nn=10)

# Get results
crf_factors = ca.crf_results
nmf_results = ca.nmf_results

Functional API (R one-to-one mirror)

from celladmix import knn_adjacency_matrix, run_knn_nmf, run_crf_all

# KNN adjacency
adj = knn_adjacency_matrix(df, k=20)

# NMF
nmf_res = run_knn_nmf(df, k=5, h=20)

# CRF segmentation
crf_res = run_crf_all(df, nmf_res, num_nn=10)

Performance Benchmarks

Correlation with R Reference

Function	Metric	Value	Gate
Gene Probabilities	Pearson r	1.00000000	PASS
Sparse Correlation	Pearson r	1.00000000	PASS
Normalize Images	Max error	0.00e+00	PASS

Speed Comparison (vs R)

Function	Python	R	Speedup
KNN Count Matrix	0.03s	~3.4s	~100x
NMF Factorization	1.24s	~21s	~17x
Gene Probabilities	0.002s	-	-
Sparse Correlation	instant	-	-

Key Optimizations

1. KNN Batch Processing: Vectorized KNN computation instead of per-cell loops (100x faster)
2. sklearn NMF: Optimized NMF implementation with multiple seeds (17x faster than R)
3. Sparse Operations: Efficient scipy.sparse operations throughout

What's included

Python function	R function
knn_adjacency_matrix	knn.adjacency.matrix
knn_count_matrix	knn.count.matrix
get_knn_counts_all	get.knn.counts.all
run_knn_nmf	run.knn.nmf
run_crf_all	run.crf.all
estimate_cell_adjacency	estimate.cell.adjacency
estimate_cell_type_adjacency	estimate.cell.type.adjacency
estimate_gene_prob_per_type	estimate.gene.prob.per.type
estimate_contamination_scores	estimate.contamination.scores
sparse_cor	sparse.cor
estimate_correlation_preservation	estimate.correlation.preservation
run_bridge_test	run.bridge.test
extract_bridge_res	extract.bridge.res
run_memb_test	run.memb.test
extract_memb_res	extract.memb.res
get_enr	get.enr
check_f_rm	check.f.rm
check_f_rm_per_ct	check.f.rm.per.ct
samp_ct_equal	samp.ct.equal
subset_genes	subset.genes
get_counts_matrix	get.counts.meta.seurat
normalize_images	normalize.images
plot_nmf_loadings	plot.nmf.loadings
plot_cell_score_ratios	plot.cell.score.ratios
plot_annot_hmap	plot.annot.hmap
plot_correlation_preservation	plot.correlation.preservation
plot_expression_comparison	plot.expression.comparison

Reproducing R results exactly

# Run the parity test
import subprocess
subprocess.run(['pytest', 'tests/test_parity.py', '-v'])

Test Results

22 passed, 4 skipped, 0 failed
├── 9 smoke tests (all pass)
├── 13 parity tests (all pass)
└── 4 exact_match tests (skipped — CRF unavailable on R 4.5)

Relationship to omicverse

py-cellAdmix is designed for use with the omicverse ecosystem.

Citation

If you use this package, please cite the original cellAdmix paper:

Mitchel et al. "Evaluating and correcting cell admixtures in imaging-based spatial transcriptomics data."

License

MIT (same as upstream cellAdmix)