Skip to contents

This page takes you from a clean R session to real results on public data. You need R 4.1.0 or later, and no Python or sign-up.

The two public datasets here let you test the software. They don’t reproduce the paper’s figures, which used a separate dataset.

1. Install

install.packages(c("remotes", "arrow"))
remotes::install_github("RaredonLab/NICHESv2", ref = "main")

2. Confirm it works

This takes a few seconds and uses data that ship with the package.

library(NICHESv2)

data(NICHESv2_inputs)
obj <- create_NICHESObject(
  count.mtx = NICHESv2_inputs$count.mtx,
  meta.data = NICHESv2_inputs$meta.data,
  LRM.db    = NICHESv2_inputs$LRM.db,
  mode      = "spatial",
  k         = 5L
)
dim(obj)

You should see 80 cells, 50 genes, 354 edges and 15 LRMs.

3. Get the test data

Both datasets are attached to a release of the NICHESv2 repository, and download_NICHESv2_data() fetches them. Each tutorial calls it for you. The downloads are about 14 MB and 11 MB, happen once, and need no other package. To fetch both now:

4. Run the tutorials

Tutorial Data Time with n.cores = 2
Spatial: mouse brain 2,696 Visium spots About 4 minutes
Non-spatial: stimulated PBMCs 13,999 cells About 1 minute, 4 GB of memory

Each tutorial’s code runs from top to bottom, so you can copy it directly into R. Results should match the printed output, because scoring is deterministic and nonspatial sampling uses a fixed seed.

For a quick check, these are the expected object sizes:

Tutorial Cells Genes Edges LRMs
Spatial 2,696 31,053 18,500 3,097
Non-spatial 13,999 14,053 87,957 1,047

The spatial tutorial stores 5,191,520 non-zero scores, and the non-spatial tutorial stores 605,973.

5. Explore further

If something goes wrong

  • A download fails: check your internet connection and try again. The files are listed on the release page.
  • A step runs out of memory: close other programs. The non-spatial tutorial needs about 4 GB.
  • A step feels slow: raise n.cores if your computer has more cores.
  • Something else: please open an issue.