Write scTransform-corrected counts to a new store — sct_corrected_counts

Reverses the residual transform with every latent variable, the library size included, held at its median. The depth structure goes, the per-sample intercept stays.

The result is a new store on disk, not a layer on object. Its gene axis is the model's, so it is narrower than the source and the indices do not line up, which is why the observation and variable tables are rebuilt rather than copied.

Usage

sct_corrected_counts_sc(
  object,
  dir_out = NULL,
  build_cell_store = TRUE,
  overwrite = FALSE,
  gene_batch_size = NULL,
  .verbose = TRUE
)

Arguments

object: SingleCells or SingleCellsSubset class with a scTransform fit attached. The analytic Pearson model has no corrected-count equivalent.
dir_out: String or NULL. Directory to write to. NULL uses sct_corrected inside the object's own data directory.
build_cell_store: Boolean. Also write the counts_cells.bin companion and the database, giving back a SingleCells rather than a path. Costs a second pass over the data.
overwrite: Boolean. Overwrite an existing store in dir_out.
gene_batch_size: Integer or NULL. Genes held in memory per batch.
.verbose: Boolean or Integer. Controls verbosity.

Value

With build_cell_store = TRUE a new SingleCells over the corrected counts, otherwise the path of the gene-major file, invisibly.

Examples

# corrected counts as a fresh object
sc <- demo_single_cells(prepped = FALSE)
sc <- fit_residuals_sc(sc, .verbose = FALSE)
corrected <- sct_corrected_counts_sc(sc, .verbose = FALSE)
dim(corrected)
#> [1] 500  50

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