generateBackgroundModel: Generate background models

In saeyslab/triwise: Gene expression changes in three biological conditions

Description

Generates a background model by randomly resampling genes at different 'n' (number of genes) and angles and calculating z distributions

Usage

generateBackgroundModel(barycoords, noi = seq(5, 100, 5),
  anglesoi = seqClosed(0, 2 * pi, 24), nsamples = 1e+05, bw = 20,
  mc.cores = getOption("mc.cores", default = 1))

Arguments

barycoords	Dataframe containing for every gene its barycentric coordinates, as returned by transformBarycentric. Will use the z column as test statistic, or if this column is not given the 'r' column
noi	Integer vector denoting the number of genes at which to sample, the larger the more accurate the p-values
anglesoi	Numeric vector denoting the angles (in radians) at which to pre-calculate null distribution, the larger the more accurate the p-values
nsamples	Number of samples, higher for more accurate and stable p-values
bw	Bandwidth of the von-mises distribution for weighing the samples. A higher bandwidth leads to a more accurate p-value estimate as long as 'nsamples' is high enough
mc.cores	Number of processor cores to use. Due to limitations of the parallel package, this does not work on Windows

Value

A list containing:

• noi: number of genes, in the same order as the elements in backmodels

• anglesoi: angles at which weights were calculated using the von-mises distribution

• nsamples: number of samples for each n

• bw: bandwidth

• backmodels: for each n a second list containing:

  • angles: mean angle of a sample

  • z: strength of unidirectional upregulation

  • weights: weight from von-mises distribution for every sample and angle in anglesoi, these weights will be used to calculate the p-value

Examples

Eoi = matrix(rnorm(1000*3, sd=0.5), 1000, 3, dimnames=list(1:1000, c(1,2,3)))
Eoi[1:100,1] = Eoi[1:100,1] + 1
barycoords = transformBarycentric(Eoi)

hist(barycoords$angle)

bm = generateBackgroundModel(barycoords)

# the distribution of mean angle of the samples is not uniform due to the
# non-uniform distribution of the angles of individual genes
hist(bm$backmodels[[1]]$angles)

# the whole distribution (and therefore also the p-value) also depends on the mean angle
plotdata = data.frame(angle = cut(bm$backmodels[[1]]$angles, 10), z = bm$backmodels[[1]]$z)
ggplot2::ggplot(plotdata) + ggplot2::geom_violin(ggplot2::aes(angle, z))

saeyslab/triwise documentation built on May 29, 2019, 12:56 p.m.