Convert Between Seurat and SingleCellExperiment

Many Bioconductor workflows – scran normalization, scater quality control, slingshot trajectory inference – operate on the SingleCellExperiment (SCE) container. Seurat provides built-in converters (as.SingleCellExperiment() and as.Seurat()) that make it straightforward to move data between the two ecosystems. scConvert extends this by letting you route SCE objects to any file format (h5ad, h5Seurat, loom, zarr, rds) through its hub architecture.

Load demo data

scConvert ships a small PBMC dataset. We start by reading it into a Seurat object.

pbmc <- readH5AD(system.file("extdata", "pbmc_demo.h5ad", package = "scConvert"))
pbmc
#> An object of class Seurat 
#> 2000 features across 500 samples within 1 assay 
#> Active assay: RNA (2000 features, 2000 variable features)
#>  2 layers present: counts, data
#>  2 dimensional reductions calculated: pca, umap

Seurat to SingleCellExperiment

Convert the Seurat object to SCE. Counts, normalized data, dimensional reductions, and cell metadata are all transferred.

library(SingleCellExperiment)

sce <- as.SingleCellExperiment(pbmc)
sce
#> class: SingleCellExperiment 
#> dim: 2000 500 
#> metadata(0):
#> assays(2): counts logcounts
#> rownames(2000): PPBP LYZ ... CLIC2 HEMGN
#> rowData names(0):
#> colnames(500): AAACATTGAGCTAC AAACATTGATCAGC ... TTTCGAACTCTCAT
#>   TTTGCATGCCTCAC
#> colData names(8): orig.ident nCount_RNA ... seurat_clusters ident
#> reducedDimNames(2): PCA UMAP
#> mainExpName: RNA
#> altExpNames(0):

Inspect what was carried over:

cat("Assays:", paste(assayNames(sce), collapse = ", "), "\n")
#> Assays: counts, logcounts
cat("Reduced dims:", paste(reducedDimNames(sce), collapse = ", "), "\n")
#> Reduced dims: PCA, UMAP
cat("colData columns:", paste(colnames(colData(sce)), collapse = ", "), "\n")
#> colData columns: orig.ident, nCount_RNA, nFeature_RNA, seurat_annotations, percent.mt, RNA_snn_res.0.5, seurat_clusters, ident

SingleCellExperiment to Seurat

Convert the SCE back to a Seurat object with as.Seurat(). The converter detects which assays exist (counts, logcounts) and maps them to the appropriate Seurat layers.

seurat_rt <- as.Seurat(sce, counts = "counts", data = "logcounts")
seurat_rt
#> An object of class Seurat 
#> 2000 features across 500 samples within 1 assay 
#> Active assay: RNA (2000 features, 0 variable features)
#>  2 layers present: counts, data
#>  2 dimensional reductions calculated: PCA, UMAP

Verify the round-trip

Confirm that expression values, cell barcodes, and embeddings survive the Seurat -> SCE -> Seurat round-trip.

# Cell barcodes
cat("Barcodes match:", identical(Cells(pbmc), Cells(seurat_rt)), "\n")
#> Barcodes match: TRUE

# Gene names
cat("Features match:", identical(rownames(pbmc), rownames(seurat_rt)), "\n")
#> Features match: TRUE

# Count matrix values
orig_counts <- GetAssayData(pbmc, layer = "counts")
rt_counts <- GetAssayData(seurat_rt, layer = "counts")
cat("Counts identical:", identical(orig_counts, rt_counts), "\n")
#> Counts identical: TRUE

# Dimensionality reductions (SCE stores these in reducedDims)
cat("Reductions in original:", paste(names(pbmc@reductions), collapse = ", "), "\n")
#> Reductions in original: pca, umap
cat("Reductions after round-trip:", paste(names(seurat_rt@reductions), collapse = ", "), "\n")
#> Reductions after round-trip: PCA, UMAP

SCE to h5ad via scConvert

To share an SCE dataset with Python collaborators, pass it directly to scConvert(). The SCE is converted to Seurat internally, then written to the target format – no manual intermediate step needed.

h5ad_path <- tempfile(fileext = ".h5ad")
scConvert(sce, dest = h5ad_path, overwrite = TRUE)
cat("h5ad file size:", round(file.size(h5ad_path) / 1e6, 1), "MB\n")
#> h5ad file size: 0.9 MB

Read it back to confirm the data survived the full SCE -> Seurat -> h5ad -> Seurat chain:

pbmc_h5ad <- readH5AD(h5ad_path)
cat("Cells:", ncol(pbmc_h5ad), "\n")
#> Cells: 500
cat("Genes:", nrow(pbmc_h5ad), "\n")
#> Genes: 2000
cat("Reductions:", paste(names(pbmc_h5ad@reductions), collapse = ", "), "\n")
#> Reductions: PCA, UMAP

Seurat to SCE via scConvert

The reverse also works. Use dest = "sce" to get a SingleCellExperiment directly from any Seurat object:

sce2 <- scConvert(pbmc, dest = "sce")
cat("Class:", class(sce2), "\n")
#> Class: SingleCellExperiment
cat("Dimensions:", nrow(sce2), "x", ncol(sce2), "\n")
#> Dimensions: 2000 x 500

Practical example: Bioconductor QC on SCE, then back to Seurat

A common workflow is to use Bioconductor tools for quality control or normalization, then bring the result back into Seurat for downstream analysis. Here we compute log-library-size and detected-feature counts using base R (the same quantities that scater::addPerCellQCMetrics would compute).

# Compute QC metrics on the SCE
sce_qc <- as.SingleCellExperiment(pbmc)
counts_mat <- assay(sce_qc, "counts")
colData(sce_qc)$log_library_size <- log1p(Matrix::colSums(counts_mat))
colData(sce_qc)$detected_features <- Matrix::colSums(counts_mat > 0)

cat("Added QC columns:",
    paste(c("log_library_size", "detected_features"), collapse = ", "), "\n")
#> Added QC columns: log_library_size, detected_features

# Convert back to Seurat -- the new colData columns become metadata
seurat_qc <- as.Seurat(sce_qc, counts = "counts", data = "logcounts")
cat("Seurat metadata columns:",
    paste(colnames(seurat_qc[[]]), collapse = ", "), "\n")
#> Seurat metadata columns: orig.ident, nCount_RNA, nFeature_RNA, seurat_annotations, percent.mt, RNA_snn_res.0.5, seurat_clusters, ident, log_library_size, detected_features

The new QC columns (log_library_size, detected_features) are now available as Seurat metadata for filtering or visualization.

Clean up

unlink(h5ad_path)