bioconductor-despace

---

name: bioconductor-despace description: Intuitive framework for identifying spatially variable genes (SVGs) via edgeR, a popular method for performing differential expression analyses. Based on pre-annotated spatial clusters as summarized spatial information, DESpace models gene expression using a negative binomial (NB), via edgeR, with spatial clusters as covariates. SVGs are then identified by testing the significance of spatial clusters. The method is flexible and robust, and is faster than the most SV methods. Furthermore, to the when_to_use: Use when: A framework to discover spatially variable genes; Differential expression analysis (DESpace); Single-cell RNA-seq analysis (DESpace). Not for: Requires biological replicates for statistical analysis; Requires R ≥ 4.3.0 and Bioconductor ≥ 3.16 user-invocable: false

DESpace

Workflows

Standard Workflow

Identify spatially variable genes (SVGs) across tissue regions globally or within individual spatial clusters for single or multiple biological replicates.

library(DESpace)
library(SpatialExperiment)

# Assuming spe3 is a SpatialExperiment object with spatial clusters in 'layer_guess_reordered'
results <- svg_test(spe = spe3, cluster_col = "layer_guess_reordered", verbose = TRUE)

# Test individual clusters using pre-computed dispersions
cluster_results <- individual_svg(spe3, edgeR_y = results$estimated_y, cluster_col = "layer_guess_reordered")

# Combine results
merge_res <- top_results(results$gene_results, cluster_results)

Inputs are a SpatialExperiment object with annotated spatial clusters; outputs are gene-level and cluster-level spatial variability statistics.

Differential Spatial Patterns

Identify genes with differential spatial expression patterns (DSPs) across multiple experimental conditions or time points.

library(DESpace)
library(SpatialExperiment)

# Assuming spe.combined is a combined SpatialExperiment object with multiple samples
multi_results <- svg_test(spe = spe.combined, 
                          cluster_col = "layer_guess_reordered", 
                          sample_col = "sample_id", 
                          replicates = TRUE)

# Test individual clusters across replicates
cluster_results <- individual_svg(spe.combined, 
                                  edgeR_y = multi_results$estimated_y, 
                                  replicates = TRUE, 
                                  cluster_col = "layer_guess_reordered")

# Combine results
merge_res <- top_results(multi_results$gene_results, cluster_results, select = "FDR")

Inputs are a combined multi-sample SpatialExperiment object; outputs are joint gene-level and cluster-level spatially variable gene statistics across replicates.

When to Use

• Identifying spatially variable genes (SVGs) from spatial transcriptomics data using svg_test.
• Testing spatial variability within individual spatial clusters using individual_svg.
• Jointly modeling multiple biological replicates to find consistent spatial patterns using replicates = TRUE in svg_test.
• Combining gene-level and cluster-level results to filter for highly or lowly abundant SVGs using top_results.

When NOT to Use

• For spatial clustering itself, use BayesSpace or stLearn because DESpace requires pre-annotated spatial clusters as input.
• For single-cell RNA-seq differential expression without spatial coordinates, use standard edgeR or DESeq2 because DESpace is designed for spatially resolved transcriptomics.

Data Requirements

• Input Format: A SpatialExperiment or SingleCellExperiment object.
• Structure: Must contain spatial coordinates (e.g., array_row, array_col) and pre-annotated spatial clusters (e.g., layer_guess_reordered) in colData.
• Normalization: Raw count matrix in the counts assay (modeled via negative binomial in edgeR).

Key Parameters

• spe: The input SpatialExperiment or SingleCellExperiment object.
• cluster_col: Column name in colData(spe) containing spatial clusters.
• sample_col: Column name in colData(spe) containing sample IDs (for multi-sample).
• replicates (FALSE): Logical indicating whether to fit the multi-sample model.
• edgeR_y: Pre-computed DGEList object containing dispersion estimates to speed up individual_svg.
• verbose (TRUE): Logical to print progress and statistics.
• high_low ("both"): Filter in top_results to select "high", "low", or "both" abundant genes.

Best Practices

• Perform quality control filtering to remove low-quality spots (e.g., low library size, high mitochondrial ratio) and lowly abundant genes (e.g., detected in < 20 spots) before testing.
• Use pre-computed gene-level dispersion estimates (estimated_y) in individual_svg to significantly speed up computation.
• Ensure cluster labels are consistent across multiple samples when performing multi-sample joint testing.
• Visualize the spatial expression of top SVGs using FeaturePlot with cluster outlines.

Common Pitfalls

• Inconsistent cluster labels across samples: Ensure that cluster names refer to the same spatial region across all samples before running multi-sample joint testing.
• Slow computation in individual cluster tests: Pass the estimated_y from svg_test to the edgeR_y parameter in individual_svg to avoid re-estimating dispersions.
• Including low-quality spots: Apply spot-level QC (e.g., using scuttle::addPerCellQC) and filter out genes with low counts before running svg_test.

Alternatives

• edgeR for non-spatial differential expression analysis.
• BayesSpace for subspot-resolution spatial clustering.
• SpatialLIBD for interactive visualization and exploration of DLPFC spatial data.

Citations

• Robinson, M. D., McCarthy, D. J., & Smyth, G. K. (2010). edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics, 26(1), 139-140.
• Maynard, K. R. et al. (2021). Transcriptome-scale spatial gene expression in the human dorsolateral prefrontal cortex. Nature Neuroscience, 24(3), 425-436.

References

• Homepage: bioconductor.org/packages/DESpace
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/DESpace/inst/doc/DESpace.html