scde.fit.models.to.reference: Fit scde models relative to provided set of expression...

In scde: Single Cell Differential Expression

Description

If group-average expression magnitudes are available (e.g. from bulk measurement), this method can be used to fit individual cell error models relative to that reference

Usage

scde.fit.models.to.reference(counts, reference, n.cores = 10,
  zero.count.threshold = 1, nrep = 1, save.plots = FALSE,
  plot.filename = "reference.model.fits.pdf", verbose = 0, min.fpm = 1)

Arguments

counts	count matrix
reference	a vector of expression magnitudes (read counts) corresponding to the rows of the count matrix
n.cores	number of cores to use
zero.count.threshold	read count to use as an initial guess for the zero threshold
nrep	number independent of mixture fit iterations to try (default = 1)
save.plots	whether to write out a pdf file showing the model fits
plot.filename	model fit pdf filename
verbose	verbose level
min.fpm	minimum reference fpm of genes that will be used to fit the models (defaults to 1). Note: fpm is calculated from the reference count vector as reference/sum(reference)*1e6

Value

matrix of scde models

Examples

data(es.mef.small)
cd <- clean.counts(es.mef.small, min.lib.size=1000, min.reads = 1, min.detected = 1)

o.ifm <- scde.error.models(counts = cd, groups = sg, n.cores = 10, threshold.segmentation = TRUE)
o.prior <- scde.expression.prior(models = o.ifm, counts = cd, length.out = 400, show.plot = FALSE)
# calculate joint posteriors across all cells
jp <- scde.posteriors(models = o.ifm, cd, o.prior, n.cores = 10, return.individual.posterior.modes = TRUE, n.randomizations = 100)
# use expected expression magnitude for each gene
av.mag <- as.numeric(jp$jp %*% as.numeric(colnames(jp$jp)))
# translate into counts
av.mag.counts <- as.integer(round(av.mag))
# now, fit alternative models using av.mag as a reference (normally this would correspond to bulk RNA expression magnitude)
ref.models <- scde.fit.models.to.reference(cd, av.mag.counts, n.cores = 1)

scde documentation built on Nov. 8, 2020, 6:19 p.m.