Convert a SingleCellExperiment assay to a scipy CSC sparse matrix

Extracts an assay from a SingleCellExperiment::SingleCellExperiment object, coerces it to a Matrix::dgCMatrix, and constructs a Python scipy.sparse.csc_matrix by passing reticulate::np_array() views of the underlying R memory (true zero-copy: flags.owndata = FALSE).

Slot mapping

R (dgCMatrix)	Python / scipy CSC	dtype
x@x	data	float64
x@i	indices	int32
x@p	indptr	int32
x@Dim	shape	–

Zero-copy semantics

reticulate::np_array() uses the C buffer protocol to expose R's contiguous memory to NumPy without copying (flags.owndata = FALSE). This works because plain R vectors are already C-contiguous:

• R numeric (double) maps directly to NumPy float64.

• R integer (int32_t) maps directly to NumPy int32.

Note on np.array(..., copy=False): NumPy >= 2.0 changed this flag to raise ValueError when a copy would be needed. Using reticulate::np_array() avoids this entirely and is the canonical reticulate API for zero-copy array creation from R.

Lifetime caveat: the returned Python object holds a borrowed reference to the underlying R vectors. Do not modify, reallocate, or allow the mat@x, mat@i, mat@p vectors to be garbage-collected while the Python object is alive.

Why CSC?

dgCMatrix stores columns contiguously (Compressed Sparse Column). We map directly to csc_matrix without reordering values. Building a CSR matrix would require transposing the data (= always a copy).

int32 for indptr

R integer vectors are 32-bit (int32_t). scipy CSC accepts int32 for both indices and indptr, supporting up to \(2^{31}-1\) non-zero entries per matrix.

Usage

sce_assay_to_scipy_csc(x, assay = "counts")

Arguments

x

A SingleCellExperiment::SingleCellExperiment.

assay

A single string: the name of the assay to extract (default "counts").

Value

A Python scipy.sparse.csc_matrix (a reticulate Python object; not auto-converted to R).

See also

sce_to_anndata(), sce_to_reticulate_anndata()

Examples

if (FALSE) { # \dontrun{
library(SingleCellExperiment)
library(Matrix)
counts <- rsparsematrix(200, 50, density = 0.1, repr = "C")
sce <- SingleCellExperiment(assays = list(counts = counts))
csc <- sce_assay_to_scipy_csc(sce, assay = "counts")
sp <- reticulate::import("scipy.sparse", convert = FALSE)
shape_r <- reticulate::py_to_r(csc$shape)
cat("Shape:", shape_r[[1]], "x", shape_r[[2]], "\n")
cat("Sparse:", reticulate::py_to_r(sp$issparse(csc)), "\n")
cat("owndata:", reticulate::py_to_r(csc$data$flags$owndata), "\n")
} # }