thresholdSCRNACountMatrix: Threshold a count matrix using an adaptive threshold.

In RGLab/MAST: Model-based Analysis of Single Cell Transcriptomics

Threshold a count matrix using an adaptive threshold.

Description

An adaptive threshold is calculated from the conditional mean of expression, based on 10 bins of the genes with similar expression levels. Thresholds are chosen by estimating cutpoints in the bimodal density estimates of the binned data. These density estimates currently exclude the zeros due to complications with how the bandwidth is selected. (If the bandwith is too small, then extra peaks/modes are found and everything goes haywire). If the diagnostic plots don't reveal any bimodal bins, this is probably the reason, and you may not need to threshold since background in the data are exact zeros.

Usage

thresholdSCRNACountMatrix(
  data_all,
  conditions = NULL,
  cutbins = NULL,
  nbins = 10,
  bin_by = "median",
  qt = 0.975,
  min_per_bin = 50,
  absolute_min = 0,
  data_log = TRUE,
  adj = 1
)

Arguments

data_all	matrix of (possibly log-transformed) counts or TPM. Rows are genes and columns are cells.
conditions	Bins are be determined per gene and per condition. Typically contrasts of interest should be specified.
cutbins	vector of cut points.
nbins	integer number of bins when cutbins is not specified.
bin_by	character "median", "proportion", "mean"
qt	when bin_by is "quantile", what quantile should be used to form the bins
min_per_bin	minimum number of genes within a bin
absolute_min	numeric giving a hard threshold below which everything is assumed to be noise
data_log	is data_all log+1 transformed? If so, it will be returned on the (log+1)-scale as well.
adj	bandwith adjustment, passed to density

Value

list of thresholded counts (on natural scale), thresholds, bins, densities estimated on each bin, and the original data

Examples

data(maits,package='MAST', envir = environment())
sca <- FromMatrix(t(maits$expressionmat[,1:1000]), maits$cdat, maits$fdat[1:1000,])
tt <- thresholdSCRNACountMatrix(assay(sca))
tt <- thresholdSCRNACountMatrix(2^assay(sca)-1, data_log=FALSE)
opar <- par(no.readonly = TRUE)
on.exit(par(opar))
par(mfrow=c(4,2))
plot(tt)

RGLab/MAST documentation built on Sept. 30, 2023, 1:08 p.m.