metaseqR (version 1.12.2)

estimate.aufc.weights: Estimate AUFC weights

Description

This function automatically estimates weights for the "weight" and "dperm.weight" options of metaseqR for combining p-values from multiple statistical tests. It creates simulated dataset based on real data and then performs statistical analysis with metaseqR several times in order to derive False Discovery Curves. Then, the average areas under the false discovery curves are used to construct weights for each algorithm, according to its performance when using simulated data.

Usage

estimate.aufc.weights(counts, normalization,
        statistics, nsim = 10, N = 10000, 
        samples = c(3, 3), ndeg = c(500, 500),
        top = 500, model.org = "mm9", fc.basis=1.5,
        seed = NULL, draw.fpc = FALSE, multic = FALSE,
        ...)

Arguments

counts

the real raw counts table from which the simulation parameters will be estimated. It must not be normalized and must contain only integer counts, without any other annotation elements and unique gene identifiers as the rownames attribute.

normalization

same as normalization in metaseqr.

statistics

same as statistics in metaseqr.

nsim

the number of simulations to perform to estimate the weights. It default to 10.

N

the number of genes to produce. See make.sim.data.sd.

samples

a vector with 2 integers, which are the number of samples for each condition (two conditions currently supported).

ndeg

a vector with 2 integers, which are the number of differentially expressed genes to be produced. The first element is the number of up-regulated genes while the second is the number of down-regulated genes.

fc.basis

the minimum fold-change for deregulation.

top

the top top best ranked (according to p-value) to use, to calculate area under the false discovery curve.

model.org

the organism from which the data are derived. It must be one of metaseqr supported organisms.

seed

a list of seed for reproducible simulations. Defaults to NULL.

draw.fpc

draw the averaged false discovery curves? Default to FALSE.

multic

whether to run in parallel (if package parallel is present or not.

...

Further arguments to be passed to estimate.sim.params.

Value

• A vector of weights to be used in metaseqr with the weights option.

Examples

data("mm9.gene.data",package="metaseqR")
multic <- check.parallel(0.8)
weights <- estimate.aufc.weights(
    counts=as.matrix(mm9.gene.counts[,9:12]),
    normalization="edaseq",
    statistics=c("deseq","edger"),
    nsim=3,N=100,ndeg=c(10,10),top=10,model.org="mm9",
    seed=10,multic=multic,libsize.gt=1e+5
)