muscat: Multi-sample multi-group scRNA-seq analysis Tools

The muscat workflows require a SingleCellExperiment object (sce) with specific columns in its cell metadata (colData). You must ensure the following identifiers are present:

1. sample_id: Unique identifiers for each sample (e.g., PeterPan1, Nautilus7).
2. group_id: Experimental condition for each sample (e.g., reference/stimulated, healthy/diseased).
3. cluster_id: Subpopulation or cluster assignments (e.g., B cells, dendritic cells).

Aggregation-based methods involve two steps: first computing pseudobulks by summing counts, then running the differential state analysis on those pseudobulks.

1. Compute pseudobulks: Use aggregateData() to sum counts by cluster_id and sample_id.
2. Run DS analysis: Use pbDS() on the resulting pseudobulk object using a specified method (e.g., "edgeR").

# compute pseudobulks (sum of counts)
pb <- aggregateData(sce, 
    assay = "counts", fun = "sum",
    by = c("cluster_id", "sample_id"))
    
# run pseudobulk (aggregation-based) DS analysis
ds_pb <- pbDS(pb, method = "edgeR")

To install muscat, use BiocManager from R. This package is available through the Bioconductor repository.

if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("muscat")

To test for differential detection (as proposed by Gilis et al., 2025), use the "num.detected" function during aggregation and then apply the differential detection testing methods.

1. Aggregate binarized counts: Use aggregateData() with fun = "num.detected".
2. Test for differential detection: Use either pbDD(pb) or pbDS(pb, method = "DD") on the aggregated object.

# sum binarized counts
pb <- aggregateData(sce, 
    assay = "counts", 
    fun = "num.detected", # !!!
    by = c("cluster_id", "sample_id"))

# test for differential detection
dd <- pbDD(pb) # or..
dd <- pbDS(pb, method = "DD")

Mixed models can be run directly on cell-level measurements without manual pseudobulk aggregation. Use mmDS() and specify a method such as "dream".

# Mixed models run directly on cell-level measurements
ds_mm <- mmDS(sce, method = "dream")