meta.perm: Permutation tests for meta-analysis

In pmoulos/metaseqr: An R package for the analysis and result reporting of RNA-Seq data by combining multiple statistical algorithms.

Description

This function performs permutation tests in order to derive a meta p-value by combining several of the statistical algorithms of metaseqr. This is probably the most accurate way of combining multiple statistical algorithms for RNA-Seq data, as this issue is different from the classic interpretation of the term "meta-analysis" which implies the application of the same statistical test on different datasets treating the same subject/experiment. For other methods, see also the main metaseqr help page. You should keep in mind that the permutation procedure can take a long time, even when executed in parallel.

Usage

    meta.perm(contrast, counts, sample.list, statistics,
        stat.args, libsize.list, nperm = 10000,
        weight = rep(1/ncol(counts), ncol(counts)),
        select = c("min", "max", "weight"), replace = "auto",
        reprod=TRUE, multic = FALSE)

Arguments

contrast	the contrasts to be tested by each statistical algorithm. See the main metaseqr help page.
counts	a normalized read counts table, one row for each gene, one column for each sample.
sample.list	the list containing condition names and the samples under each condition. See the main metaseqr help page.
statistics	the statistical algorithms used in metaseqr. See the main metaseqr help page.
stat.args	the parameters for each statistical algorithm. See the main metaseqr help page.
libsize.list	a list with library sizes. See the main metaseqr and the stat.* help pages.
nperm	the number of permutations (Monte Carlo simulations) to perform.
weight	a numeric vector of weights for each statistical algorithm.
select	how to select the initial vector of p-values. It can be "min" to select the minimum p-value for each gene (more conservative), "max" to select the maximum p-value for each gene (less conservative), "weight" to apply the weights to the p-value vector for each gene and derive a weighted p-value.
replace	same as the replace argument in the sample function. Implies bootstraping or simple resampling without replacement. It can also be "auto", to determine bootstraping or not with the following rule: if ncol(counts)<=6 replace=FALSE else replace=TRUE. This protects from the case of having zero variability across resampled conditions. In such cases, most statistical tests would crash.
reprod	create reproducible permutations. Ideally one would want to create the same set of indices for a given dataset so as to create reproducible p-values. If reprod=TRUE, a fixed seed is used by meta.perm for all the datasets analyzed with metaseqr. If reprod=FALSE, then the p-values will not be reproducible, although statistical significance is not expected to change for a large number of resambling. Finally, reprod can be a numeric vector of seeds with the same length as nperm so that the user can supply his/her own seeds.
multic	use multiple cores to execute the premutations. This is an external parameter and implies the existence of multicore package in the execution environment. See the main metaseqr help page.

Value

A vector of meta p-values

Author(s)

Panagiotis Moulos

Examples

# Not yet available

pmoulos/metaseqr documentation built on Dec. 29, 2020, 5:56 a.m.