3 Coarse-graining of large single-cell data into metacells using SuperCell

Here, we use the pbmc4k dataset as presented in Batch correction.

3.1 RunSuperCell

The RunSuperCell() function is a wrapper function to run SuperCell.

pbmc <- RunSuperCell(pbmc4k, gamma = 5)
pbmc

## class: SingleCellExperiment 
## dim: 33694 842 
## metadata(3): hvgmethod hvgcols SuperCell
## assays(2): counts logcounts
## rownames(33694): ENSG00000243485 ENSG00000237613 ...
##   ENSG00000277475 ENSG00000268674
## rowData names(7): ENSEMBL_ID Symbol_TENx ...
##   variance.expected variance.standardized
## colnames: NULL
## colData names(1): size
## reducedDimNames(0):
## mainExpName: NULL
## altExpNames(0):

The output of RunSuperCell() is a SingleCellExperiment object that stores gene expression matrix of metacells.

We can use it for downstram analysis.

post_process <- function(pbmc) {
    pbmc <- NormalizeData(pbmc)
    pbmc <- FindVariableFeatures(pbmc)
    pbmc <- ScaleData(pbmc)
    pbmc <- runPCA(pbmc, subset_row = VariableFeatures(pbmc), exprs_values = "scaled")
    
    pbmc <- FindNeighbors(pbmc, dims = 1:10)
    pbmc <- FindClusters(pbmc)
    pbmc <- RunUMAP(pbmc)
    return(pbmc)
}

pbmc2 <- post_process(pbmc)

library(ggsc)
sc_dim(pbmc2, reduction="UMAP") + sc_dim_geom_label()

3.2 Estimate SuperCell purity

pbmc4k2 <- post_process(pbmc4k)

SC <- metadata(pbmc2)$SuperCell

purity <- SuperCell::supercell_purity(pbmc4k2$label, SC$membership, method = 'entropy')

head(purity)

##         1         2         3         4         5         6 
## 0.0000000 1.0397208 0.4505612 1.0549202 0.3046361 0.5004024

summary(purity)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##  0.0000  0.0000  0.0000  0.1805  0.4506  1.3863