selectBatchHVG: Batch-aware highly variable gene selection

Description

Method to select HVGs based on mean dispersions of genes that are highly variable genes in all batches. Using a the top target_genes per batch by average normalize dispersion. If target genes still hasn't been reached, then HVGs in all but one batches are used to fill up. This is continued until HVGs in a single batch are considered.

This is an rliger implementation of the method originally published in SCIB. We found the potential that it can improve integration under some circumstances, and is currently testing it.

This function currently only works for shared features across all datasets. For selection from only part of the datasets and selection for dataset-specific unshared features, please use selectGenes().

Usage

selectBatchHVG(object, ...)
# S3 method for liger
selectBatchHVG(
  object,
  nGenes = 2000,
  verbose = getOption("ligerVerbose", TRUE),
  ...
)
# S3 method for ligerDataset
selectBatchHVG(
  object,
  nGenes = 2000,
  features = NULL,
  scaleFactor = NULL,
  verbose = getOption("ligerVerbose", TRUE),
  ...
)
# S3 method for dgCMatrix
selectBatchHVG(
  object,
  nGenes = 2000,
  returnStats = FALSE,
  scaleFactor = NULL,
  verbose = getOption("ligerVerbose", TRUE),
  ...
)
# S3 method for DelayedArray
selectBatchHVG(
  object,
  nGenes = 2000,
  means = NULL,
  scaleFactor = NULL,
  returnStats = FALSE,
  chunk = getOption("ligerChunkSize", 20000),
  verbose = getOption("ligerVerbose", TRUE),
  ...
)

Value

liger-method: Returns the input liger object with the selected genes updated in varFeatures slot, which can be accessed with varFeatures(object). Additionally, the statistics are updated in the featureMeta slot of each ligerDataset object within the datasets slot of the object.
ligerDataset-method: Returns the input ligerDataset object with the statistics updated in the featureMeta slot.
dgCMatrix-method: By default returns a character vector of selected variable features. If returnStats = TRUE, returns a data.frame of the statistics.

Arguments

object: A liger object, ligerDataset object or a sparse/dense matrix. The liger objects must have raw counts available. A direct matrix input is preferably log-1p transformed from CPM normalized counts in cell per column orientation.
...: Arguments passed to S3 methods.
nGenes: Integer number of target genes to select. Default 2000.
verbose: Logical. Whether to show a progress bar. Default getOption("ligerVerbose") or TRUE if users have not set.
features: For ligerDataset method, the feature subset to limit the selection to, mainly for limiting the selection to happen within the shared genes of all datasets. Default NULL selects from all features in the ligerDataset object.
scaleFactor: Numeric vector of scaling factor to normalize the raw counts to unit sum. This pre-calculated at liger object creation (stored as object$nUMI and internally specified in S3 method chains, thus is generally not needed to be specified by users.
returnStats: Logical, for dgCMatrix-method, whether to return a data frame of statistics for all features, or by default FALSE just return a character vector of selected features.
means: Numeric vector of pre-calculated means per gene, derived from log1p CPM normalized expression.
chunk: Integer. Number of maximum number of cells in each chunk when working on HDF5Array Default 20000.

References

Luecken, M.D., Büttner, M., Chaichoompu, K. et al. (2022), Benchmarking atlas-level data integration in single-cell genomics. Nat Methods, 19, 41–50. https://doi.org/10.1038/s41592-021-01336-8.

Examples

Run this code

pbmc <- selectBatchHVG(pbmc, nGenes = 10)
varFeatures(pbmc)