Python port of SpaceTrooper — Quality Control for image-based spatial transcriptomics data.
GitHub: github.com/omicverse/py-SpaceTrooper | PyPI: py-spacetrooper
pip install py-spacetrooper import anndata as ad
from spacetrooper import SpaceTrooper
# Load your spatial data
adata = ad .read_h5ad ("your_data.h5ad" )
# Run QC pipeline
st = SpaceTrooper (adata )
st .spatial_per_cell_qc ()
st .compute_qc_score ()
# Access results
print (st .adata .obs ["QC_score" ]) # QC scores (0-1) from spacetrooper import (
spatial_per_cell_qc ,
compute_outliers_qc_score ,
check_outliers ,
compute_qc_score ,
)
spatial_per_cell_qc (adata )
compute_outliers_qc_score (adata )
check_outliers (adata )
compute_qc_score (adata )
Platform
Function
Description
Nanostring CosMx
read_cosmx_speTranscriptomics + Proteomics
Vizgen MERFISH
read_merfish_speMERFISH / Merscope
10x Xenium
read_xenium_speXenium output bundle
Benchmarked on the SpaceTrooper bundled CosMx dataset (905 cells, 1010 genes) against R reference output:
QC_score (primary output)
Metric
Value
Gate
Status
Pearson r 0.989 ≥ 0.90
✅
Max absolute error
0.194
—
—
Mean absolute error
0.027
—
—
R mean
0.809
—
—
Python mean
0.755
—
—
Intermediate metrics (machine-precision match)
Metric
Max abs error
Gate
Status
log2SignalDensity
4.88e-15
< 1e-8
✅
Area_um
5.68e-14
< 1e-8
✅
total
0.00e+00
< 1e-8
✅
ctrl_total_ratio
~1e-15
< 1e-8
✅
Outlier labels (exact match)
Label
R
Python
log2SignalDensity LOW
38
38 ✅
log2SignalDensity HIGH
19
19 ✅
Area_um HIGH
6
6 ✅
log2AspectRatio LOW
5
5 ✅
log2AspectRatio HIGH
11
11 ✅
log2Ctrl_total_ratio HIGH
9
9 ✅
Full QC pipeline (spatialPerCellQC → computeOutliersQCScore → checkOutliers → computeQCScore) on 905-cell CosMx data:
Platform
Wall-clock
Speedup
Python (py-spacetrooper) 0.068s 102×
R (SpaceTrooper)
6.956s
1×
R ⇄ Python function dictionary
R function
Python function
Notes
spatialPerCellQC(spe, micronConvFact=0.12, rmZeros=TRUE, negProbList=...)spatial_per_cell_qc(adata, micron_conv_fact=0.12, rm_zeros=True, neg_prob_list=...)SPE → AnnData
computeThresholdFlags(spe, totalThreshold=0, ctrlTotRatioThreshold=0.1)compute_threshold_flags(adata, total_threshold=0, ctrl_tot_ratio_threshold=0.1)—
R function
Python function
Notes
computeSpatialOutlier(spe, computeBy, method="mc", mcDoScale=FALSE, scuttleType="both")compute_spatial_outlier(adata, compute_by, method="mc", mc_do_scale=False, scuttle_type="both", nmads=3)Added nmads param
computeOutliersQCScore(spe, metricList=...)compute_outliers_qc_score(adata, metric_list=...)—
checkOutliers(spe, verbose=FALSE)check_outliers(adata, verbose=False)—
getFencesOutlier(spe, fencesOf, highLow="both", decimalRound=NULL)get_fences_outlier(adata, fences_of, high_low="both", decimal_round=None)Thresholds in adata.uns
R function
Python function
Notes
computeQCScore(spe, bestLambda=NULL, verbose=FALSE)compute_qc_score(adata, best_lambda=None, verbose=False)sklearn replaces glmnet
computeTrainDF(colData, formulaVars, tech, verbose=FALSE)compute_train_df(coldata, formula_vars, tech, verbose=False)—
computeLambda(trainDF, modelFormula)compute_lambda(train_df, var_names)Returns fixed lambda=7.9
trainModel(modelMatrix, trainDF)train_model(model_matrix, train_df, C=None)Returns (model, mean, std)
getModelFormula(formulaVars, verbose=FALSE)get_model_formula(formula_vars, verbose=False)Returns variable name list
R function
Python function
Notes
readPolygons(polygonsFile, type="csv", x=..., y=..., xloc=..., yloc=...)read_polygons(polygons_file, file_type="csv", x=..., y=..., xloc=..., yloc=...)sf → GeoDataFrame
readPolygonsCosmx(polygonsFile, type="csv", ...)read_polygons_cosmx(polygons_file, file_type="csv", ...)—
readPolygonsXenium(polygonsFile, type="parquet", ...)read_polygons_xenium(polygons_file, file_type="parquet", ...)—
readPolygonsMerfish(polygons, type="parquet", ...)read_polygons_merfish(polygons_source, file_type="parquet", ...)HDF5 not implemented
computeAreaFromPolygons(polygons)compute_area_from_polygons(gdf)—
computeAspectRatioFromPolygons(polygons)compute_aspect_ratio_from_polygons(gdf)—
computeCenterFromPolygons(polygons, coldata)compute_center_from_polygons(gdf, coldata)—
addPolygonsToSPE(spe, polygons, polygonsCol="polygons")add_polygons_to_adata(adata, polygons, polygons_col="polygons")SPE → AnnData
R function
Python function
Notes
readCosmxSPE(dirName, sampleName="sample01", ...)read_cosmx_spe(dir_name, sample_name="sample01", ...)Returns AnnData
readMerfishSPE(dirName, sampleName="sample01", ...)read_merfish_spe(dir_name, sample_name="sample01", ...)Returns AnnData
readXeniumSPE(dirName, sampleName="sample01", ...)read_xenium_spe(dir_name, sample_name="sample01", ...)Returns AnnData
R function
Python function
Notes
plotMetricHist(spe, metric, fillColor=..., useFences=..., fencesColors=...)plot_metric_hist(adata, metric, fill_color=..., use_fences=..., fences_colors=...)ggplot2 → matplotlib
plotCentroids(spe, colourBy=..., pointCol=..., size=...)plot_centroids(adata, colour_by=..., point_color=..., size=...)ggplot2 → matplotlib
plotQScoreTerms(spe, sampleId=..., size=...)plot_qc_score_terms(adata, sample_id=..., size=...)ggplot2 → matplotlib
R function
Python function
Notes
.getActiveGeometryName(sf)get_active_geometry_name(gdf)sf → GeoDataFrame
.setActiveGeometry(sf, name)set_active_geometry(gdf, name)—
.renameGeometry(sf, from, to, activate=FALSE)rename_geometry(gdf, from_name, to_name, activate=False)—
R concept
Python equivalent
SpatialExperimentAnnData
sf (simple features)GeoDataFrame (geopandas)
ggplot2matplotlib + seaborn
glmnet(alpha=0)sklearn.LogisticRegression(penalty='l2')
scuttle::isOutlier(nmads=3)_is_outlier_scuttle(nmads=3)
colData(spe)adata.obs
metadata(spe)adata.uns
spatialCoords(spe)adata.obsm['spatial']
assays(spe)$countsadata.X
@article {righelli2024spacetrooper ,
title ={ SpaceTrooper: Quality Control for image-based spatial transcriptomics} author ={ Righelli, Dario and Banzi, Benedetta and Marchionni, Matteo and others} year ={ 2024} journal ={ Bioinformatics} MIT (matches upstream SpaceTrooper )