Spatially-Aware Differential Expression

Pipeline question: Which genes are differentially expressed across spatial contexts — between spatial domains, near specific cell types, or along spatial gradients?

Overview

Spatially-aware differential expression (DE) extends standard DE testing by incorporating spatial information. Rather than comparing predefined groups (as in bulk or scRNA-seq DE), spatial DE tests whether gene expression changes as a function of spatial position, neighborhood composition, or proximity to specific structures. This is distinct from spatially variable gene detection (which asks "is this gene spatially patterned?") — spatial DE asks "is this gene differentially expressed between spatial contexts?"

Key Methods

C-SIDE (spacexr)

• Paper: Nature Biotechnology, 2022
• Code: github.com/dmcable/spacexr
• Key innovation: Cell-type-specific spatially-informed differential expression within the spacexr framework — tests for DE while accounting for cell-type mixtures in each spot.
• Strengths:
  • Handles cell-type mixtures natively (uses RCTD deconvolution)
  • Tests for cell-type-specific spatial expression changes
  • Well-integrated with RCTD deconvolution in a single package
• Limitations:
  • Requires prior deconvolution via RCTD
  • R only
  • Designed for spot-level (Visium) data
• Technology compatibility: Visium, Slide-seq

Niche-DE

• Paper: Nature Genetics, 2024
• Code: github.com/KChen-lab/Niche-DE
• Key innovation: Tests whether genes in a focal cell type are differentially expressed depending on the cell-type composition of the spatial neighborhood (niche).
• Strengths:
  • Directly tests the niche hypothesis — does neighborhood composition affect expression?
  • Identifies niche-associated genes per cell type
  • Handles both spot-level and cell-level data
• Limitations:
  • Requires accurate cell-type annotations or deconvolution
  • Statistical power depends on sufficient niche variation
• Technology compatibility: Visium, MERFISH, Xenium, any spatial platform with cell-type labels

Vespucci

• Paper: bioRxiv, 2024
• Code: github.com/smiranast/Vespucci
• Key innovation: Framework for DE analysis that models gene expression as a function of spatial covariates, enabling flexible spatial regression testing.
• Strengths:
  • Flexible covariate-based testing (spatial coordinates, distance to structure, etc.)
  • Can test complex spatial hypotheses
• Limitations:
  • Requires careful specification of spatial covariates
  • Preprint with limited community adoption
• Technology compatibility: Visium, Slide-seq

CSDE

• Paper: Bioinformatics, 2023
• Code: github.com/Uauy-Lab/CSDE
• Key innovation: Cell-type-specific differential expression that decomposes spot-level changes into cell-type-specific components without requiring single-cell resolution.
• Strengths:
  • Attributes spot-level DE to specific cell types
  • Works with spot-level data where individual cells are not resolved
• Limitations:
  • Accuracy depends on deconvolution quality
  • Limited validation on imaging-based platforms
• Technology compatibility: Visium, Slide-seq

spatialGE

• Paper: Bioinformatics, 2024
• Code: github.com/FridleyLab/spatialGE
• Key innovation: R toolkit for spatial gene expression analysis that includes spatial DE testing alongside visualization and spatial statistics.
• Strengths:
  • Integrated toolkit covering multiple spatial analysis tasks
  • Good visualization capabilities
  • Designed for clinical/translational spatial studies
• Limitations:
  • Jack-of-all-trades — DE component is not as specialized as dedicated methods
  • R only
• Technology compatibility: Visium, CosMx, any spatial platform

Benchmark Summary

No systematic benchmark currently compares spatial DE methods head-to-head. C-SIDE is the most established method through its integration with RCTD/spacexr, providing a natural deconvolution-to-DE pipeline for Visium data. Niche-DE addresses a unique and important question — niche-associated expression changes — that other methods do not directly test. For standard two-group spatial DE (e.g., tumor vs. stroma), Wilcoxon or edgeR/DESeq2 applied to spatial domains remains a practical and valid approach.

Standard DE tools still work

For comparing predefined spatial regions (e.g., tumor core vs. margin), standard scRNA-seq DE tools (Wilcoxon, edgeR, DESeq2) applied to spatial domain labels are effective. Spatial DE methods add value when the question involves gradients, niches, or cell-type-specific spatial effects.

Spatial autocorrelation inflates significance

Standard DE tests assume independent observations. Spatially adjacent spots are correlated, inflating test statistics and producing false positives. Methods like C-SIDE and Niche-DE account for this; standard tests applied to spatial data should use spatial bootstrap or downsampling for calibration.

When to Use What

Your data	Your goal	Recommended	Why
Visium with RCTD deconvolution	Cell-type-specific spatial DE	C-SIDE	Integrated deconvolution-to-DE pipeline
Cell-resolution spatial	Test niche effects on expression	Niche-DE	Directly tests neighborhood composition effects
Any spatial data, two groups	Compare spatial regions	Standard DE (Wilcoxon/edgeR)	Simple, effective for discrete region comparisons
Spot-level, want cell-type DE	Decompose spot-level changes	CSDE	Attributes changes to cell types in mixtures

Technology Compatibility

Method	Visium	Visium HD	Xenium	MERFISH	CosMx	CODEX	Stereo-seq
C-SIDE	Yes	-	-	-	-	-	-
Niche-DE	Yes	-	Yes	Yes	Yes	-	-
Vespucci	Yes	-	-	-	-	-	-
CSDE	Yes	-	-	-	-	-	-
spatialGE	Yes	-	-	-	Yes	-	-