Skip to contents

Impute a subset of genes with MAGIC

run_magic_sc.Rd

MAGIC smooths counts over the kNN graph: a row-stochastic diffusion operator is built from the graph stored on the object and applied n_steps times, so every cell becomes a weighted average of its neighbourhood. It makes sparse marker genes readable on an embedding and gives gene trends something less speckled to fit. For details, please refer to van Dijk, et al.

Deliberately restricted to a gene subset. The output is dense and the binary store is never touched, so pick the genes you want to plot or fit and impute those. Genome-wide imputation is not offered.

Each call replaces the whole layer. Read it back with get_magic(), or hand it to any of the extractors with layer = "magic".

Usage

run_magic_sc(
  object,
  features,
  modality = c("rna", "adt", "wnn"),
  magic_params = params_sc_magic(),
  .verbose = TRUE
)

Arguments

object

One of SingleCells, SingleCellsSubset or SingleCellsMultiModal. MetaCells are already aggregated and are not supported.

features

Character vector. The genes to impute. Anything not found in the object is dropped with a warning.

modality

String. One of c("rna", "adt", "wnn"). Which kNN graph does the smoothing. Anything but "rna" requires a SingleCellsMultiModal object. The counts themselves always come from the RNA store, and the layer is always written to the RNA cache.

magic_params

List. See params_sc_magic().

.verbose

Boolean or integer. Controls verbosity and returns run times. FALSE -> quiet, TRUE or 1L -> normal verbosity, 2L -> detailed verbosity.

Value

The object with an ScMagic layer in its cache.

What this does to your results

Neighbourhoods overlap, so smoothing over them makes nearby cells resemble each other in every gene at once. Gene-gene correlation after imputation is inflated, badly, and the inflation is a property of the graph rather than of the biology. Do not feed imputed values into Hotspot, SCENIC, differential correlation or CoReMo: those methods measure exactly the quantity MAGIC manufactures. Visualisation and trend fitting are what it is for.

The operator preserves per-cell mass, so imputed values sit on the scale of whatever went in. Imputing raw counts and imputing log-normalised counts are different operations rather than the same one rescaled, see the layer element of params_sc_magic().

References

van Dijk, et al., Cell, 2018.

Examples

# diffusion imputation of five genes for plotting
sc <- demo_single_cells()
sc <- run_magic_sc(
  sc,
  features = get_gene_names(sc)[1:5],
  .verbose = FALSE
)
dim(get_magic(sc)$data)
#> [1] 500   5

unlink(sc@dir_data, recursive = TRUE, force = TRUE)