AnnData-Conversion: Convert AnnData between and SingleCellExperiment
In zellkonverter: Conversion Between scRNA-seq Objects

Description Usage Arguments Details Value Author(s) See Also Examples

Conversion between Python AnnData objects and SingleCellExperiment objects.

1
2
3

AnnData2SCE(adata, skip_assays = FALSE, hdf5_backed = TRUE)

SCE2AnnData(sce, X_name = NULL, skip_assays = FALSE)

`adata`	A reticulate reference to a Python AnnData object.
`skip_assays`	Logical scalar indicating whether to skip conversion of any assays in `sce` or `adata`, replacing them with empty sparse matrices instead.
`hdf5_backed`	Logical scalar indicating whether HDF5-backed matrices in `adata` should be represented as HDF5Array objects. This assumes that `adata` is created with `backed="r"`.
`sce`	A SingleCellExperiment object.
`X_name`	Name of the assay to use as the primary matrix (`X`) of the AnnData object. If `NULL`, the first assay of `sce` will be used by default.

These functions assume that an appropriate Python environment has already been loaded. As such, they are largely intended for developer use, most typically inside a basilisk context.

The conversion is not entirely lossless. The current mapping is shown below (also at https://tinyurl.com/AnnData2SCE):

SCE-AnnData map

In SCE2AnnData(), matrices are converted to a numpy-friendly format. Sparse matrices are converted to dgCMatrix objects while all other matrices are converted into ordinary matrices. If skip_assays = TRUE, empty sparse matrices are created instead and the user is expected to fill in the assays on the Python side.

For AnnData2SCE(), a warning is raised if there is no corresponding R format for a matrix in the AnnData object, and an empty sparse matrix is created instead as a placeholder. If skip_assays = NA, no warning is emitted but variables are created in the int_metadata() of the output to specify which assays were skipped. If skip_assays = TRUE, empty sparse matrices are created for all assays, regardless of whether they might be convertible to an R format or not. In both cases, the user is expected to fill in the assays on the R side, see readH5AD() for an example.

We attempt to convert between items in the SingleCellExperiment metadata() slot and the AnnData uns slot. If an item cannot be converted a warning will be raised.

Values stored in the varm slot of an AnnData object are stored in a column of rowData() in a SingleCellExperiment as a DataFrame of matrices. No attempt is made to transfer this information when converting from SingleCellExperiment to AnnData.

AnnData2SCE() will return a SingleCellExperiment containing the equivalent data from adata.

SCE2AnnData() will return a reticulate reference to an AnnData object containing the content of sce.

Luke Zappia

Aaron Lun

writeH5AD() and readH5AD() for dealing directly with H5AD files.

if (requireNamespace("scRNAseq", quietly = TRUE)) {
    library(basilisk)
    library(scRNAseq)
    seger <- SegerstolpePancreasData()

    # These functions are designed to be run inside
    # a specified Python environment
    roundtrip <- basiliskRun(fun = function(sce) {
        # Convert SCE to AnnData:
        adata <- SCE2AnnData(sce)

        # Maybe do some work in Python on 'adata':
        # BLAH BLAH BLAH

        # Convert back to an SCE:
        AnnData2SCE(adata)
    }, env = zellkonverter:::anndata_env, sce = seger)
}