Fast identification of over-expressed genes and ligand-receptor (L-R)
interactions in single-cell RNA-seq data, working directly on
SingleCellExperiment / SummarizedExperiment objects. The per-gene,
per-cell-group Wilcoxon tests are implemented in Rust
(extendr) and run in parallel on all CPU cores.
Analysts of single-cell RNA-seq data who want to find which L-R interactions are supported by over-expressed genes in their cell groups, either as a quick standalone screen or as a faster pre-filter before a full cell-cell communication analysis (e.g. with CellChat).
cellchatr reimplements CellChat's pre-processing steps (gene subsetting, over-expressed gene detection, L-R matching), but it does not depend on CellChat and does not use CellChat objects:
SingleCellExperiment, SummarizedExperiment or a gene × cell
matrix.ligand, receptor and
interaction_name columns. CellChat's CellChatDB.human$interaction has
this layout and can be passed in as-is, but CellChat is optional.DESCRIPTION because it is distributed only on
GitHub.if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("cellchatr")
Building from source requires Rust (cargo, rustc).
library(cellchatr)
# sce: a SingleCellExperiment with a "logcounts" assay and a
# "cell_type" column in colData
# lrdb: a data.frame with ligand / receptor / interaction_name columns,
# e.g. CellChatDB.human$interaction
res <- identifyInteractions(sce, lrdb,
labels = "cell_type",
assay.type = "logcounts",
pval_threshold = 0.05)
res # DataFrame of matched interactions
metadata(res)$overexpressed # over-expressed genes
Or step by step:
oe <- findOverexpressedGenes(sce, labels = "cell_type", lrdb = lrdb)
matchLRPairs(oe, lrdb)
See vignette("cellchatr-workflow", package = "cellchatr") for a full
example.
| Function | Input | Output |
|---|---|---|
identifyInteractions() |
SingleCellExperiment / SummarizedExperiment / matrix + L-R table |
DataFrame of interactions |
findOverexpressedGenes() |
SingleCellExperiment / SummarizedExperiment / matrix (dense or sparse) |
over-expressed genes |
matchLRPairs() |
genes + L-R table |
DataFrame of interactions |
rust_wilcoxon_filter(), rust_subset_genes(), rust_match_lr_pairs()
|
plain vectors / matrix | low-level Rust routines |
Statistics. One-sided Wilcoxon rank-sum test (group vs. all other cells,
alternative "greater"), normal approximation with tie correction and no
continuity correction, equivalent to
wilcox.test(..., alternative = "greater", exact = FALSE, correct = FALSE).
A gene is reported if it is significant in at least one group.
The Wilcoxon step is where the time goes: one test per gene per cell group.
cellchatr runs these tests in parallel in Rust, and the vignette includes a
reproducible timing comparison against an equivalent wilcox.test() loop.
For the simple lookups, base R's %in% is already fast, so the Rust versions
are not meant as speed-ups.
Under review: Bioconductor/Contributions #137
MIT