runSims: Run a number of RNA-seq simulations and DE detections
In haowulab/PROPER: PROspective Power Evaluation for RNAseq

Description Usage Arguments Details Value Author(s) See Also Examples

View source: R/simfuns.R

This is the "wrapper" function for running RNA-seq DE detection simulation. It runs simulations under different sample sizes (replicates in each group) for a certain numbers. In each simulation, the RNA-seq data are generated, and then DE detection (using user specified method/software) is performed. The return object contains DE test results (test statistics, p-values, FDRs, etc.) from all simulations.

1 2	runSims(Nreps = c(3, 5, 7, 10), Nreps2, nsims = 100, sim.opts, DEmethod = c("edgeR", "DSS", "DESeq2"), verbose =TRUE)

`Nreps`	Sample sizes one wants to perform simulation on. This is a vector, each element specifies the number of biological replicates in each group. Default value is c(3, 5, 7, 10), which means we want to perform simulation when there are 3, 5, 7, or 10 replicates in each group.
`Nreps2`	Sample sizes for the second treatment group. If this is missing, it'll take the same value as Nreps and then two groups will be assumed to have the same sample size. This parameter allows two treatment groups have different sample sizes. When specified, it must be a vector of the same length as Nreps.
`nsims`	Number of simulations.
`sim.opts`	An object for simulation option. This should be the return object from "RNAseq.SimOptions.2grp" function.
`DEmethod`	String to specify the DE detection method to be used. Available options are "edgeR", "DSS", and "DESeq2".
`verbose`	Logical value to indicate whether to output some messages (progress report).

This is the main simulation function in the packge. After specifying the simulation parameters (from "RNAseq.SimOptions.2grp" function), one wants to evaluate the power vs different sample sizes under that simulation setting. This function simulates the count data and performs statistical tests for DE detection. It only stores and returns the DE test results (test statistics, p-values, FDRs, etc.) but doesn't make inferences. The inferences will be conducted in the "comparePower" function. The advantage is that for one simulation setting, the simulation only need to be run once. The inferences using different critical values and type I error controls can then be drawn from the same results. This greatly save the computation because the simulation part is the most computationally intensive.

This function can be slow, depends on the setting (number of genes, replicates, simulations, etc). For the default (50000 genes, 100 simulations, for 3, 5, 7, or 10 replicates), it takes about an hour to run on a single core of i7 2.7GHz CPU. But again, this only need to be run once for a particular simulation setting.

A list with following fields:

`pvalue, fdrs`	3D array for p-values and FDR from each simulation. The dimension of the array is ngenes * N * nsims. Here N is length(Nreps), of the number of different sample sizes settings.
`xbar`	3D array for average read counts for genes. Dimension is the same as pvalue/fdr. This will be used in "comparePower" function for stratified power quantities.
`DEid`	A list of length nsims. Each contains the id of DE genes.
`lfcs`	A list of length nsims. Each contains the log fold changes of DE genes.
`Nreps`	The input Nreps.
`sim.opts`	The input sim.opts.

Hao Wu <hao.wu@emory.edu>

RNAseq.SimOptions.2grp, simRNAseq

simOptions = RNAseq.SimOptions.2grp()
## using 3 different sample sizes, run 2 simulations, using edgeR
simRes = runSims(Nreps=c(3,5,7), sim.opts=simOptions, nsims=2,
                 DEmethod="edgeR")
names(simRes)

## Not run: 
## using 5 different sample sizes, run 100 simulations, using edgeR.
## This will be slow.
simRes = runSims(Nreps=c(3,5,7,10,15), sim.opts=simOptions, nsims=100,
                 DEmethod="edgeR")

## for different sample sizes in two groups
simRes = runSims(Nreps=c(3,5,7,10), Nreps2=c(5,7,9,11),
                 sim.opts=simOptions, nsims=100, DEmethod="edgeR")


## End(Not run)