cellMarkers: Identify cell markers

In cellGeometry: Geometric Single Cell Deconvolution

Identify cell markers

Description

Uses geometric method based on vector dot product to identify genes which are the best markers for individual cell types.

Usage

cellMarkers(
  scdata,
  bulkdata = NULL,
  subclass,
  cellgroup = NULL,
  nsubclass = 25,
  ngroup = 10,
  expfilter = 0.5,
  noisefilter = 2,
  noisefraction = 0.25,
  min_cells = 10,
  remove_subclass = NULL,
  dual_mean = FALSE,
  meanFUN = "logmean",
  postFUN = NULL,
  verbose = TRUE,
  sliceMem = 16,
  cores = 1L,
  ...
)

Arguments

scdata	Single-cell data matrix with genes in rows and cells in columns. Can be sparse matrix or DelayedMatrix. Must have rownames representing gene IDs or gene symbols.
bulkdata	Optional data matrix containing bulk RNA-Seq data with genes in rows and samples in columns. This matrix is only used for its rownames (gene IDs), to ensure that cell markers are selected from genes in the bulk dataset.
subclass	Vector of cell subclasses matching the columns in scdata
cellgroup	Optional grouping vector of major cell types matching the columns in scdata. subclass is assumed to contain subclasses which are subsets within cellgroup overarching classes.
nsubclass	Number of genes to select for each single cell subclass. Either a single number or a vector with the number of genes for each subclass.
ngroup	Number of genes to select for each cell group. Either a single number or a vector with the number of genes for each group.
expfilter	Genes whose maximum mean expression on log2 scale per cell type are below this value are removed and not considered for the signature.
noisefilter	Sets an upper bound for noisefraction cut-off below which gene expression is set to 0. Essentially gene expression above this level must be retained in the signature. Setting this higher can allow more suppression via noisefraction and can favour more highly expressed genes.
noisefraction	Numeric value. Maximum mean log2 gene expression across cell types is calculated and values in celltypes below this fraction are set to 0. Set in conjunction with noisefilter. Note: if this is set too high (too close to 1), it can have a deleterious effect on deconvolution.
min_cells	Numeric value specifying minimum number of cells in a subclass category. Subclass categories with fewer cells will be ignored.
remove_subclass	Character vector of subclass levels to be removed from the analysis.
dual_mean	Logical whether to calculate arithmetic mean of counts as well as mean(log2(counts +1)). This is mainly useful for simulation.
meanFUN	Either a character value or function for applying mean which is passed to scmean(). Options include "logmean" (the default) or "trimmean" which is a trimmed after excluding the top/bottom 5% of values. rowMeans can be used for the arithmetic mean of counts.
postFUN	Optional function applied to genemeans matrices after mean has been calculated. If meanFUN is set to "trimmean" or "rowMeans", then postFUN defaults to log2s. See scmean().
verbose	Logical whether to show messages.
sliceMem	Max amount of memory in GB to allow for each subsetted count matrix object. When scdata is subsetted by each cell subclass, if the amount of memory would be above sliceMem then slicing is activated and the subsetted count matrix is divided into chunks and processed separately. This is indicated by addition of '...' in the printed timings. The limit is just under 17.2 GB (2^34 / 1e9). Above this the subsetted matrix breaches the long vector limit (>2^31 elements).
cores	Integer, number of cores to use for parallelisation using mclapply(). Parallelisation is not available on windows. Warning: parallelisation has increased memory requirements. See scmean().
...	Additional arguments passed to scmean() such as use_future.

Details

If verbose = TRUE, the function will display an estimate of the required memory. But importantly this estimate is only a guide. It is provided to help users choose the optimal number of cores during parallelisation. Real memory usage might well be more, theoretically up to double this amount, due to R's use of copy-on-modify.

Value

A list object with S3 class 'cellMarkers' containing:

call	the matched call
best_angle	named list containing a matrix for each cell type with genes in rows. Rows are ranked by lowest specificity angle for that cell type and highest maximum expression. Columns are: angle the specificity angle in radians, angle.deg the same angle in degrees, max the maximum mean expression across all cell types, rank the rank of the mean gene expression for that cell type compared to the other cell types
group_angle	named list of matrices similar to best_angle, for each cell subclass
geneset	character vector of selected gene markers for cell types
group_geneset	character vector of selected gene markers for cell subclasses
genemeans	matrix of mean log2+1 gene expression with genes in rows and cell types in columns
genemeans_filtered	matrix of gene expression for cell types following noise reduction
groupmeans	matrix of mean log2+1 gene expression with genes in rows and cell subclasses in columns
groupmeans_filtered	matrix of gene expression for cell subclasses following noise reduction
cell_table	factor encoded vector containing the groupings of the cell types within cell subclasses, determined by which subclass contains the maximum number of cells for each cell type
spillover	matrix of spillover values between cell types
subclass_table	contingency table of the number of cells in each subclass
opt	list storing options, namely arguments nsubclass, ngroup, expfilter, noisefilter, noisefraction
genemeans_ar	if dual_mean is TRUE, optional matrix of arithmetic mean, i.e. log2(mean(counts)+1)
genemeans_filtered_ar	optional matrix of arithmetic mean following noise reduction

The 'cellMarkers' object is designed to be passed to deconvolute() to deconvolute bulk RNA-Seq data. It can be updated rapidly with different settings using updateMarkers(). Ensembl gene ids can be substituted for recognisable gene symbols by applying gene2symbol().

Author(s)

Myles Lewis

See Also

deconvolute() updateMarkers() gene2symbol()

cellGeometry documentation built on April 20, 2026, 1:06 a.m.