samr.assess.samplesize {samr}

Assess the sample size for a SAM analysis

Description

Estimate the false discovery rate, false negative rate, power and type I error for a SAM analysis. Currently implemented only for two class (unpaired or paired), one-sample and survival problems).

Usage

samr.assess.samplesize(samr.obj, data, dif, samplesize.factors=c(1,2,3,5), 
min.genes = 10, max.genes = nrow(data$x)/2)

Arguments

samr.obj

Object returned from call to samr

data

Data list, same as that passed to samr.train

dif

Change in gene expression between groups 1 and 2, for genes that are differentially expressed. For log base 2 data, a value of 1 means a 2-fold change. For One-sample problems, dif is the number of units away from zero for differentially expressed genes. For survival data, dif is the numerator of the Cox score statistic (this info is provided in the output of samr).

samplesize.factors

Integer vector of length 4, indicating the sample sizes to be examined. The values are factors that multiply the original sample size. So the value 1 means a sample size of ncol(data$x), 2 means a sample size of ncol(data$x), etc.

min.genes

Minimum number of genes that are assumed to truly changed in the population

max.genes

Maximum number of genes that are assumed to truly changed in the population

Details

Estimates false discovery rate, false negative rate, power and type I error for a SAM analysis. The argument samplesize.factor allows the use to assess the effect of varying the sample size (total number of samples). A detailed description of this calculation is given in the SAM manual at http://www-stat.stanford.edu/~tibs/SAM

Value

A list with components

Results

A matrix with columns: number of genes- both the number differentially expressed genes in the population and number called significant; cutpoint- the threshold used for the absolute SAM score d; FDR, 1-power- the median false discovery rate, also equal to the power for each gene; FDR-90perc- the upper 90th percentile of the FDR; FNR, Type 1 error- the false negative rate, also equal to the type I error for each gene; FNR-90perc- the upper 90th percentile of the FNR

dif.call

Change in gene expression between groups 1 and 2, that was provided in the call to samr.assess.samplesize

difm

The average difference in SAM score d for the genes differentially expressed vs unexpressed

samplesize.factor

The samplesize.factor that was passed to samr.assess.samplesiz

n

Number of samples in input data (i.e. ncol of x component in data)

Author(s)

Jun Li and Balasubrimanian Narasimhan and Robert Tibshirani

References

Tusher, V., Tibshirani, R. and Chu, G. (2001): Significance analysis of microarrays applied to the ionizing radiation response" PNAS 2001 98: 5116-5121, (Apr 24). http://www-stat.stanford.edu/~tibs/sam

Taylor, J., Tibshirani, R. and Efron. B. (2005). The “Miss rate” for the analysis of gene expression data. Biostatistics 2005 6(1):111-117.

A more complete description is given in the SAM manual at http://www-stat.stanford.edu/~tibs/SAM

Examples

#generate some example data
set.seed(100)
x<-matrix(rnorm(1000*20),ncol=20)
dd<-sample(1:1000,size=100)

u<-matrix(2*rnorm(100),ncol=10,nrow=100)
x[dd,11:20]<-x[dd,11:20]+u

y<-c(rep(1,10),rep(2,10))

data=list(x=x,y=y, geneid=as.character(1:nrow(x)),
genenames=paste("g",as.character(1:nrow(x)),sep=""), logged2=TRUE)

log2=function(x){log(x)/log(2)}

# run SAM first
samr.obj<-samr(data,  resp.type="Two class unpaired", nperms=100)

# assess current sample size (20), assuming 1.5fold difference on log base 2 scale

samr.assess.samplesize.obj<- samr.assess.samplesize(samr.obj, data, log2(1.5))

# assess the effect of doubling the sample size

samr.assess.samplesize.obj2<- samr.assess.samplesize(samr.obj, data, log2(1.5))