Version Compatibility
Reference examples tested with: Space Ranger 4.1+ (Visium HD; nucleus/cell segmentation in the count pipeline since v4.0), scanpy 1.10+, squidpy 1.3+, spatialdata-io current (imaging platforms), matplotlib 3.8+, numpy 1.26+
Before using code patterns, verify installed versions match. If versions differ:
- Python:
pip show <package> then help(module.function) to check signatures
If code throws ImportError, AttributeError, or TypeError, introspect the installed
package and adapt the example to match the actual API rather than retrying.
Note: squidpy.read provides visium/vizgen/nanostring only — there is NO sq.read.xenium; imaging platforms load via spatialdata_io (returns a SpatialData object preserving the molecule table). Visium HD default bin is 8 µm. Confirm in-tool before quoting.
Spatial Transcriptomics Pipeline
"Analyze my spatial transcriptomics data end-to-end" -> Orchestrate data loading (squidpy/scanpy), QC, normalization, spatial neighbor analysis, spatial statistics, spatial domain detection, and tissue visualization. Composition estimation (deconvolution, spatial-deconvolution) and cell-cell communication (spatial-communication) are deliberately separate steps -- this pipeline hands off to those skills rather than inlining them.
This is a workflow skill: it owns the chaining decisions and hand-offs, not the internals of any one step.
Made-once commitments
| Commitment | Consequence inherited downstream |
|---|
| Platform class (imaging vs sequencing) | EVERY downstream choice: segment-vs-deconvolve, discovery-vs-classification, QC floors, panel-bounded-vs-whole-transcriptome |
| Coordinate system + image-registration frame | All spatial neighbors/overlays/niches; a wrong registration frame silently misplaces every spot relative to histology |
| Panel identity (targeted vs whole-transcriptome; FFPE probe vs FF poly-A) | What "gene absent" means: on a targeted panel absence = "not in panel", not "not expressed"; RIN (FF) vs DV200 (FFPE) QC metric switch |
| Spot/bin geometry (Visium 55 µm >> cell; Visium HD 2/8 µm; Xenium single-molecule) | Whether to DECONVOLVE (spot >> cell), SEGMENT/bin-up (spot << cell), or neither |
| Segmentation policy (imaging: Baysor / Cellpose / vendor Xenium; Visium HD bin-to-cell: Space Ranger v4+) | Every cell x gene value; segmentation is the DOMINANT imaging error source and over-expansion manufactures cross-type DE |
The platform-class fork (decide first)
This pipeline branches on platform class before any step. Sequencing/capture data (Visium, Visium HD, Slide-seq, Stereo-seq) are spot/bin MIXTURES of cells: QC on spot counts, normalize knowing that library size partly carries cellularity, then DECONVOLVE composition (spatial-deconvolution) rather than read a spot as one cell. Imaging/in-situ data (Xenium, MERFISH, CosMx) are single molecules: SEGMENT cells first (image-analysis), apply low-count-aware QC floors (an scRNA min_counts=500 deletes nearly every real imaging cell, whose vector is tens-to-low-hundreds of transcripts), drop on negative-control probe rate, and SKIP deconvolution. The Squidpy+Scanpy path below is written for Visium; the imaging branch is flagged at each step.
Workflow Overview
Spatial data (Space Ranger output)
|
v
[1. Load Data] ---------> Read Visium/Xenium
|
v
[2. QC & Preprocessing] -> Filter, normalize
|
v
[3. Clustering] --------> Standard scRNA-seq clustering
|
v
[4. Spatial Analysis] --> Neighbors, statistics
|
v
[5. Domain Detection] --> Spatial domains
|
v
[6. Visualization] -----> Spatial plots
|
v
Annotated spatial data
Primary Path: Squidpy + Scanpy
Step 1: Load Data
import scanpy as sc
import squidpy as sq
import numpy as np
import matplotlib.pyplot as plt
adata = sq.read.visium('spaceranger_output/')
print(f'Loaded: {adata.n_obs} spots/cells, {adata.n_vars} genes')
Step 2: Quality Control
has_mito = adata.var_names.str.startswith('MT-').any()
if has_mito:
adata.var['mt'] = adata.var_names.str.startswith('MT-')
sc.pp.calculate_qc_metrics(adata, qc_vars=['mt'], inplace=True)
else:
sc.pp.calculate_qc_metrics(adata, inplace=True)
sc.pl.spatial(adata, color='total_counts', show=False)
plt.savefig('qc_spatial.pdf')
sc.pp.filter_cells(adata, min_counts=500)
sc.pp.filter_genes(adata, min_cells=10)
if has_mito:
adata = adata[adata.obs.pct_counts_mt < 25, :]
print(f'After QC: {adata.n_obs} spots/cells')
Step 3: Normalization and Clustering
adata.layers['counts'] = adata.X.copy()
sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata, n_top_genes=2000)
adata.raw = adata
adata = adata[:, adata.var.highly_variable]
sc.pp.scale(adata, max_value=10)
sc.tl.pca(adata, n_comps=50)
sc.pp.neighbors(adata, n_neighbors=15, n_pcs=30)
sc.tl.umap(adata)
sc.tl.leiden(adata, resolution=0.5, flavor='igraph', n_iterations=2, directed=False)
sc.pl.spatial(adata, color='leiden', spot_size=1.5)
plt.savefig('clusters_spatial.pdf')
Step 4: Spatial Analysis
sq.gr.spatial_neighbors(adata, coord_type='grid', n_neighs=6)
sq.gr.nhood_enrichment(adata, cluster_key='leiden')
sq.pl.nhood_enrichment(adata, cluster_key='leiden')
plt.savefig('nhood_enrichment.pdf')
sq.gr.co_occurrence(adata, cluster_key='leiden')
sq.pl.co_occurrence(adata, cluster_key='leiden')
plt.savefig('co_occurrence.pdf')
sq.gr.spatial_autocorr(adata, mode='moran', n_perms=100, n_jobs=4)
moran = adata.uns['moranI']
svg = moran[moran['pval_norm_fdr_bh'] < 0.05].sort_values('I', ascending=False)
print('Spatially autocorrelated genes (FDR<0.05):', svg.head(10).index.tolist())
Step 5: Domain Detection
sq.gr.spatial_neighbors(adata, coord_type='grid', n_neighs=6)
sc.tl.leiden(adata, resolution=0.3, key_added='spatial_domains',
adjacency=adata.obsp['spatial_connectivities'],
flavor='igraph', n_iterations=2, directed=False)
sc.pl.spatial(adata, color='spatial_domains', spot_size=1.5)
plt.savefig('spatial_domains.pdf')
sc.pl.spatial(adata, color=['leiden', 'spatial_domains'], ncols=2)
plt.savefig('clusters_comparison.pdf')
Step 6: Visualization
genes = ['EPCAM', 'VIM', 'PTPRC', 'COL1A1']
sc.pl.spatial(adata, color=genes, ncols=2, spot_size=1.5, cmap='viridis')
plt.savefig('marker_genes_spatial.pdf')
sc.tl.rank_genes_groups(adata, 'leiden', method='wilcoxon')
sc.pl.rank_genes_groups_dotplot(adata, n_genes=5)
plt.savefig('cluster_markers.pdf')
adata.write('spatial_analyzed.h5ad')
Complete Workflow Script
import scanpy as sc
import squidpy as sq
import matplotlib.pyplot as plt
import os
data_dir = 'spaceranger_output'
output_dir = 'spatial_results'
os.makedirs(output_dir, exist_ok=True)
os.makedirs(f'{output_dir}/plots', exist_ok=True)
print('Loading data...')
adata = sq.read.visium(data_dir)
print(f'Loaded: {adata.n_obs} spots, {adata.n_vars} genes')
print('QC filtering...')
has_mito = adata.var_names.str.startswith('MT-').any()
if has_mito:
adata.var['mt'] = adata.var_names.str.startswith('MT-')
sc.pp.calculate_qc_metrics(adata, qc_vars=['mt'], inplace=True)
else:
sc.pp.calculate_qc_metrics(adata, inplace=True)
sc.pp.filter_cells(adata, min_counts=500)
sc.pp.filter_genes(adata, min_cells=10)
if has_mito:
adata = adata[adata.obs.pct_counts_mt < 25, :]
print(f'After QC: {adata.n_obs} spots')
print('Processing...')
adata.layers['counts'] = adata.X.copy()
sc.pp.normalize_total(adata, target_sum=)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata, n_top_genes=)
adata.raw = adata
adata = adata[:, adata.var.highly_variable]
sc.pp.scale(adata, max_value=)
sc.tl.pca(adata, n_comps=)
sc.pp.neighbors(adata, n_neighbors=, n_pcs=)
sc.tl.leiden(adata, resolution=, flavor=, n_iterations=, directed=)
()
sq.gr.spatial_neighbors(adata, coord_type=, n_neighs=)
sq.gr.nhood_enrichment(adata, cluster_key=)
sq.gr.spatial_autocorr(adata, mode=, n_perms=)
()
sc.pl.spatial(adata, color=, spot_size=, save=)
sq.pl.nhood_enrichment(adata, cluster_key=, save=)
adata.write()
()
Common Errors
| Symptom | Cause | Fix |
|---|
| Spot clusters mislabeled as cell types | Skipped deconvolution on multi-cell Visium spots | Deconvolve against an annotated scRNA reference; clusters = niches (spatial-deconvolution) |
| Nearly all imaging cells filtered; small cells lost | Applied scRNA QC floor (min_counts=500) to single-molecule data | Low-count-aware floors (~10 transcripts) + negative-control-probe gating |
| Spurious cross-type DE (neuronal markers in astrocytes) | Over-aggressive segmentation expansion | Molecule-aware (Baysor) or uniform re-segmentation; segmentation is critical |
| "Spatial" neighbors are wrong | Built the neighbor graph on the expression embedding | Build on PHYSICAL coordinates (grid for Visium, kNN for imaging) |
| Overlays/niches misplaced | Image-vs-expression coordinate/registration mismatch | Verify fiducial registration; keep tissue and matrix coordinates reconciled |
| "Novel cell state" on a targeted panel | Treated a fixed panel as discovery | Classification/label-transfer only; absence = not-in-panel |
| Top-Moran gene over-interpreted as regulation | Gated on raw Moran's I / read a composition marker as within-type | Gate SVGs on FDR; a top-Moran gene usually marks a spatially-clustered cell TYPE |
References
- Palla G, Spitzer H, Klein M, et al (2022) Squidpy: a scalable framework for spatial omics analysis. Nature Methods 19:171-178. DOI 10.1038/s41592-021-01358-2. (spatial neighbor graph / neighborhood enrichment / autocorrelation.)
- Kleshchevnikov V, Shmatko A, Dann E, et al (2022) Cell2location maps fine-grained cell types in spatial transcriptomics. Nature Biotechnology 40:661-671. DOI 10.1038/s41587-021-01139-4. (deconvolution: spot != cell.)
- Cable DM, Murray E, Zou LS, et al (2022) Robust decomposition of cell type mixtures in spatial transcriptomics (RCTD). Nature Biotechnology 40:517-526. DOI 10.1038/s41587-021-00830-w.
- Petukhov V, Xu RJ, Soldatov RA, et al (2022) Cell segmentation in imaging-based spatial transcriptomics with Baysor. Nature Biotechnology 40:345-354. DOI 10.1038/s41587-021-01044-w. (segmentation as the dominant imaging error source.)
Related Skills
- spatial-transcriptomics/spatial-data-io - Loading formats (Visium/imaging; SpatialData)
- spatial-transcriptomics/spatial-preprocessing - QC floors and normalization by platform
- spatial-transcriptomics/image-analysis - Cell segmentation for imaging platforms
- spatial-transcriptomics/spatial-neighbors - Physical-space neighbor graphs
- spatial-transcriptomics/spatial-statistics - Moran's I, co-occurrence, neighborhood enrichment nulls
- spatial-transcriptomics/spatial-domains - BANKSY/BayesSpace/STAGATE domain methods
- spatial-transcriptomics/spatial-deconvolution - Cell-type composition of multi-cell spots
- spatial-transcriptomics/spatial-communication - Cell-cell communication / ligand-receptor (separate hand-off)
- spatial-transcriptomics/spatial-visualization - Spatial overlays and figures
- workflows/scrnaseq-pipeline - Upstream: provides the annotated scRNA reference for deconvolution
