spiDE

spiDE identifies context-specific, neighbourhood-dependent differential expression in spatial transcriptomics data. Within an index cell type, it tests how gene expression changes with an experimental condition as a function of the local density (the niche) of surrounding cell types.

The method:

1. builds per-cell niche covariates from Gaussian kernel density estimates of each cell type at multiple spatial bandwidths (buildNiches());
2. fits a per-gene negative binomial GLM over a design containing the three-way cell type : condition : niche interactions, using the SpaNorm fitNB() engine (fitSpiDE());
3. tests the neighbourhood interactions with Wald statistics combined across correlated covariates (Brown’s method) and bandwidths (Cauchy combination), under a hierarchical (gene → index cell type → niche cell type) FDR (testSpiDE()).

Installation

spiDE depends on SpaNorm (>= 1.7.4), which exposes the negative binomial fitting engine (fitNB) and the calculateMu / invert_mat helpers.

# install.packages("BiocManager")
BiocManager::install("bhuvad/spiDE")

Quick start

library(spiDE)
data(toySpiDE)

res <- spiDE(toySpiDE, condition = "condition", covariates = "Age")
head(results(res))

See the vignette (vignette("spiDE")) for a full walk-through.