Skip to contents

Most single-cell data in R lives in Seurat objects. This page shows how extract_NICHESInputs_Seurat() turns two real Seurat objects into NICHESv2 inputs. They are the same two datasets the tutorials use, and these are the steps that made the tutorial inputs.

The objects come from SeuratData, which the Satija Lab maintains. The tutorials don’t need it, because they download the finished inputs with download_NICHESv2_data().

Install SeuratData

SeuratData isn’t on CRAN. Run this once; the downloads are about 143 MB for stxBrain and 413 MB for ifnb. SeuratData also installs the Seurat package, so the first install can take a while.

install.packages(c("remotes", "SeuratObject"))
remotes::install_github("satijalab/seurat-data")
SeuratData::InstallData(c("stxBrain", "ifnb"))

SeuratData objects keep the Seurat version they were saved with. extract_NICHESInputs_Seurat() reads both V4 and V5 objects, so you don’t need to update them first. The spatial object below is V5 and the non-spatial one is V4.

A spatial object: Visium mouse brain

stxBrain holds four sections, and the spatial tutorial uses anterior1.

data("anterior1")

spatial <- extract_NICHESInputs_Seurat(
  seurat.obj = anterior1,
  assay      = "Spatial",
  layer      = "counts",
  verbose    = FALSE
)
dim(spatial$count.mtx)
#> [1] 31053  2696

Spatial coordinates sit in the image part of the Seurat object, so add them to the metadata as x and y. Matching by barcode keeps every spot aligned.

coords <- GetTissueCoordinates(anterior1)
spatial$meta.data$x <- coords[rownames(spatial$meta.data), "x"]
spatial$meta.data$y <- coords[rownames(spatial$meta.data), "y"]
head(spatial$meta.data[, c("x", "y")], 3)
x y
AAACAAGTATCTCCCA-1 7,475 8,501
AAACACCAATAACTGC-1 8,553 2,788
AAACAGAGCGACTCCT-1 3,164 7,950

A non-spatial object: stimulated PBMCs

data("ifnb")

nonspatial <- extract_NICHESInputs_Seurat(
  seurat.obj = ifnb,
  assay      = "RNA",
  layer      = "counts",
  verbose    = FALSE
)
dim(nonspatial$count.mtx)
#> [1] 14053 13999
table(nonspatial$meta.data$stim)
#> 
#> CTRL STIM 
#> 6548 7451

No coordinates are needed here, because non-spatial mode pairs cells by sampling within cell types.

Next steps

spatial and nonspatial hold the same inputs that download_NICHESv2_data() returns. Pass either to create_NICHESObject() and continue with the spatial tutorial or the non-spatial tutorial. For other Seurat options, such as choosing a layer, see Input formats.