AnnData/Seurat Import • spacedeconv

Setup

The preferred single-cell reference data input format of spacedeconv is SingleCellExperiment. Other common single-cell formats are anndata and Seurat. This vignette shows how to use H5AD anndata files and Seurat objects as input for the spacedeconv pipeline.

For anndata, we use the zellkonverter package which is also available via conda.

Here, we set up reticulate before loading the spacedeconv and zellkonverter libraries to avoid reticulate conflicts between the two packages.

library(reticulate)
use_condaenv("spacedeconv-env", required = TRUE)
invisible(py_config())

library(spacedeconv)
library(zellkonverter)
library(Seurat)
library(SingleCellExperiment)

Load Example Data

First, we load the included example datasets.

data("single_cell_data_3")
data("spatial_data_3")

SingleCellExperiment <-> AnnData Conversion

zellkonverter allows us to convert SingleCellExperiment objects to anndata and store them as H5AD files: zellkonverter::writeH5AD()

writeH5AD(single_cell_data_3, "single_cell_data_3.h5ad")

In case we have the single-cell data only as H5AD anndata file, we can use zellkonverter to read it into R as SingleCellExperiment object: zellkonverter::readH5AD()

That way it can be used as input for the spacedeconv pipeline. In this case we read back the example data file we just wrote to H5AD.

single_cell_data_3 <- readH5AD("single_cell_data_3.h5ad", use_hdf5 = FALSE)

SingleCellExperiment <-> Seurat Conversion

Seurat allows us to convert SingleCellExperiment to Seurat objects: Seurat::as.Seurat()

single_cell_data_3 <- as.Seurat(single_cell_data_3, counts = "counts", data = NULL)

If you have single-cell data as Seurat object, you can convert it to a SingleCellExperiment to use it as input for spacedeconv: Seurat::as.SingleCellExperiment()

single_cell_data_3 <- as.SingleCellExperiment(single_cell_data_3, assay = "originalexp")

Spacedeconv Clustering Pipeline

Now we can continue with the spacedeconv pipeline as usual. In this case we will perform the clustering procedure as an example.

single_cell_data_3 <- spacedeconv::preprocess(single_cell_data_3)
spatial_data_3 <- spacedeconv::preprocess(spatial_data_3)

spatial_data_3 <- spacedeconv::normalize(spatial_data_3, method = "cpm")

signature <- spacedeconv::build_model(
  single_cell_obj = single_cell_data_3,
  cell_type_col   = "celltype_major",
  method          = "spatialdwls",
  verbose         = TRUE
)
#> class of selected matrix:  dgCMatrix 
#> [1] "finished runPCA_factominer, method == factominer"

deconv <- spacedeconv::deconvolute(
  spatial_obj     = spatial_data_3,
  single_cell_obj = single_cell_data_3,
  cell_type_col   = "celltype_major",
  method          = "spatialdwls",
  signature       = signature,
  assay_sp        = "cpm"
)
#> class of selected matrix:  dgCMatrix 
#> [1] "finished runPCA_factominer, method == factominer"

clustered <- spacedeconv::cluster(deconv, data = "expression", clusres = 0.5)

spacedeconv::plot_spatial(
  spe     = clustered,
  result  = "cluster_expression_res_0.5",
  title   = "Clustering",
  density = FALSE
)