tibs.power: tibs.power

In chr1swallace/TibsPower: Power Calculations for Microarray Expts using the Tibshirani (2006) Method

Description

Estimate power for microarray studies by resampling existing dataset

Usage

tibs.power(norm, ngenes.diff, fold.diff, nsim, sample.sizes,
  p.thr = numeric(0), n.thr = numeric(0),
  sample.sizes.stage2 = numeric(0), p.thr.stage2 = numeric(0),
  p.stage2.method = c("bonf", "fixed"))

Arguments

norm	matrix of normalized gene expression values from a reference expt
ngenes.diff	number of differentially expressed genes
fold.diff	fold difference at differentially expressed genes
nsim	number of simulations
sample.sizes	number of pairs of samples in simulated experiment, can be a vector
p.thr	p value threshold for calling significance, can be a vector to specify multiple values. You should use only one of p.thr and n.thr.
n.thr	threshold for calling the top n genes significant, can be a vector to specify multiple values. You should use only one of p.thr and n.thr.
sample.sizes.stage2	number of pairs of samples in an optional stage 2 of the simulated experiment, can be a vector to specify multiple values.
p.thr.stage2	p value threshold for calling significance in an optional stage 2 of the simulated experiment, can NOT be a vector
p.stage2.method	by default, the p.thr.stage2 will be adjusted using Bonferroni conditional on the number of genes rejected at stage one. Set p.stage2.method="fixed" to use this p.thr.stage2 unadjusted.

Details

Uses Tibshirani method (2006).

Note that this function uses mclapply to parallelise sampling. Setting 'options(mc.cores=...)' can help speed things up if you have multiple cores available. See ?mclapply for details.

Value

list summarising the median, 5th and 95th centiles of the false discovery rate (FDR) and false negative rate (FNR)

Author(s)

Chris Wallace

References

Tibshirani, R. (2006). A simple method for assessing sample sizes in microarray experiments. BMC Bioinformatics 7, 106.

Examples

## Suppose we have a reference expression dataset
norm <- matrix(rnorm(2000),200,10)
## We want to examine power for the following scenario:
tibs.power(norm,
ngenes.diff=10, # 10 genes are truly differentially expressed
fold.diff=2, # fold difference is 2 at those differentially expressed genes
nsim=10, # average 10 simulated experiments (in reality, should use many more)
sample.sizes=c(4,8), # first stage will use either 4 or 8 pairs
n.thr=50, # take the top 50 genes through to stage 2
sample.sizes.stage2=c(20,5), # stage 2 will use either 20 or 5 pairs
p.thr.stage2=0.05, # alpha at stage 2 is 0.05 ...
p.stage2.method="bonf") # ... and is corrected for the number of genes tested (50) by Bonferroni

chr1swallace/TibsPower documentation built on May 13, 2019, 6:18 p.m.