Meta-analysis for MicroArrays

Description

Combines either p-values or moderated effect sizes from different studies to find differentially expressed genes.

Details

pvalcombination and EScombination are the most important functions to combine unpaired data.

pvalcombination combines p-values from individual studies.

EScombination combines effect sizes from individual studies.

pvalcombination.paired and EScombination.paired are to be used for paired data.

IDDIDR can help in the interpretation of gain and loss of information due to meta-analysis.

Author(s)

Guillemette Marot <guillemette.marot@inria.fr>

References

Marot, G., Foulley, J.-L., Mayer, C.-D., Jaffrezic, F. (2009) Moderated effect size and p-value combinations for microarray meta-analyses. Bioinformatics. 25 (20): 2692-2699.

Examples

library(metaMA)
data(Singhdata)
EScombination(esets=Singhdata$esets,classes=Singhdata$classes)
pvalcombination(esets=Singhdata$esets,classes=Singhdata$classes)
#more details are provided in the vignette; only open it in interactive R sessions
if(interactive()){
  vignette("metaMA")
  }

Effect size combination for unpaired data

Description

Calculates effect sizes from unpaired data either from classical or moderated t-tests (Limma, SMVar) for each study and combines these effect sizes.

Usage

EScombination(esets, classes, moderated = c("limma", "SMVar", "t")[1], BHth = 0.05)

Arguments

Value

List

Note

While the invisible object resulting from this function contains the values described previously, other quantities of interest are printed: DE,IDD,Loss,IDR,IRR. All these quantities are defined in function IDDIDR and in (Marot et al., 2009)

Author(s)

Guillemette Marot

References

Marot, G., Foulley, J.-L., Mayer, C.-D., Jaffrezic, F. (2009) Moderated effect size and p-value combinations for microarray meta-analyses. Bioinformatics. 25 (20): 2692-2699.

Examples

data(Singhdata)
#Meta-analysis
res=EScombination(esets=Singhdata$esets,classes=Singhdata$classes)
#Number of differentially expressed genes in the meta-analysis
length(res$Meta)
#To plot an histogram of raw p-values
rawpval=2*(1-pnorm(abs(res$TestStatistic)))
hist(rawpval,nclass=100)

Effect size combination for paired data

Description

Calculates effect sizes from paired data either from classical or moderated t-tests (Limma, SMVar) for each study and combines these effect sizes.

Usage

EScombination.paired(logratios, moderated = c("limma", "SMVar", "t")[1], BHth = 0.05)

Arguments

Value

List

Note

While the invisible object resulting from this function contains the values described previously, other quantities of interest are printed: DE,IDD,Loss,IDR,IRR. All these quantities are defined in function IDDIDR and in (Marot et al., 2009)

Author(s)

Guillemette Marot

References

Marot, G., Foulley, J.-L., Mayer, C.-D., Jaffrezic, F. (2009) Moderated effect size and p-value combinations for microarray meta-analyses. Bioinformatics. 25 (20): 2692-2699.

Examples

data(Singhdata)
#create artificially paired data:
artificialdata=lapply(Singhdata$esets,FUN=function(x) (x[,1:10]-x[,11:20]))
#Meta-analysis
res=EScombination.paired(artificialdata)
#Number of differentially expressed genes in the meta-analysis
length(res$Meta)
#To plot an histogram of raw p-values
rawpval=2*(1-pnorm(abs(res$TestStatistic)))
hist(rawpval,nclass=100)

Integration-driven Discovery and Integration-driven Revision Rates

Description

Calculates the gain or the loss of differentially expressed genes due to meta-analysis compared to individual studies.

Usage

IDDIRR(finalde, deindst)

Arguments

Value

Author(s)

Guillemette Marot

References

Marot, G., Foulley, J.-L., Mayer, C.-D., Jaffrezic, F. (2009) Moderated effect size and p-value combinations for microarray meta-analyses. Bioinformatics. 25 (20): 2692-2699.

Examples

data(Singhdata)
out=EScombination(esets=Singhdata$esets,classes=Singhdata$classes)
IDDIRR(out$Meta,out$AllIndStudies)

## The function is currently defined as
#function(finalde,deindst)
#{
#DE=length(finalde)
#gains=finalde[which(!(finalde %in% deindst))]
#IDD=length(gains)
#IDR=IDD/DE*100
#perte=which(!(deindst %in% finalde))
#Loss=length(perte)
#IRR=Loss/length(deindst)*100
#res=c(DE,IDD,Loss,IDR,IRR)
#names(res)=c("DE","IDD","Loss","IDR","IRR")
#res
#}

Singh dataset

Description

Publicly available microarray dataset artificially split in 5 studies

Usage

data(Singhdata)

Format

List of 3 elements:

esets

List of 5 data frames corresponding to 5 artificial studies, each with 12625 genes and 20 replicates (10 normal samples and 10 tumoral samples)

classes

List of 5 numeric vectors with class memberships, one per study

geneNames

Factor with 12625 levels corresponding to gene names

Source

These data are available on the website http://www.bioinf.ucd.ie/people/ian/. We considered 50 normal samples and 50 tumoral samples, leaving out the 2 last tumoral samples. Data are already normalized.

References

Singh, D., Febbo, P. G., Ross, K., Jackson, D. G., Manola, J., Ladd, C., Tamayo, P., Renshaw, A. A., D'Amico, A. V., Richie, J. P., Lander, E. S., Loda, M., Kantoff, P.W., Golub, T. R., and Sellers,W. R. (2002). Gene expression correlates of clinical prostate cancer behavior. Cancer Cell, 1(2). 203:209.

Examples

data(Singhdata)

Empirical Bayes statistics from limma analysis with unpaired data

Description

Computes empirical Bayes statistics from limma analysis with only one group effect.

Usage

calcfit2Diffrep(C1, C2)

Arguments

Details

Returns fit2 described in limma vignette. To be used with unpaired data.

Value

fit2

Note

see Bioconductor limma vignette

Direct effect size combination

Description

Combines effect sizes already calculated.

Usage

directEScombi(ES, varES, BHth = 0.05, useREM = TRUE)

Arguments

Details

Combines effect sizes with the method presented in (Choi et al., 2003).

Value

List

References

Choi, J. K., Yu, U., Kim, S., and Yoo, O. J. (2003). Combining multiple microarray studies and modeling interstudy variation. Bioinformatics, 19 Suppl 1.

Direct p-value combination

Description

Combines one sided p-values with the inverse normal method.

Usage

directpvalcombi(pvalonesided, nrep, BHth = 0.05)

Arguments

Value

List

Note

One-sided p-values are required to avoid directional conflicts. Then a two-sided test is performed to find differentially expressed genes.

Author(s)

Guillemette Marot

References

Hedges, L. and Olkin, I. (1985). Statistical Methods for Meta-Analysis. Academic Press.

Calculates effect sizes from given t or moderated t statistics

Description

Function not to be used separately.

Usage

effectsize(tstat, ntilde, m)

Arguments

Value

Matrix with one row per gene, and in column:

Author(s)

Guillemette Marot with contribution from Ankur Ravinarayana Chakravarthy

References

Marot, G., Foulley, J.-L., Mayer, C.-D., Jaffrezic, F. Moderated effect size combination for microarray meta-analyses and comparison study. Submitted.

Examples

#for SMVar: 
#stati$TestStat[order(stati$GeneId)],length(classes[[i]]),stati$DegOfFreedom[order(stati$GeneId)])
#for Limma
#effectsize(fit2i$t,length(classes[[i]]),(fit2i$df.prior+fit2i$df.residual))

P-value combination for unpaired data

Description

Calculates differential expression p-values from unpaired data either from classical or moderated t-tests (Limma, SMVar) for each study and combines these p-values by the inverse normal method.

Usage

pvalcombination(esets, classes, moderated = c("limma", "SMVar", "t")[1], BHth = 0.05)

Arguments

Value

List

Note

While the invisible object resulting from this function contains the values described previously, other quantities of interest are printed: DE,IDD,Loss,IDR,IRR. All these quantities are defined in function IDDIDR and in (Marot et al., 2009)

Author(s)

Guillemette Marot

References

Marot, G., Foulley, J.-L., Mayer, C.-D., Jaffrezic, F. (2009) Moderated effect size and p-value combinations for microarray meta-analyses. Bioinformatics. 25 (20): 2692-2699.

Examples

data(Singhdata)
#Meta-analysis
res=pvalcombination(esets=Singhdata$esets,classes=Singhdata$classes)
#Number of differentially expressed genes in the meta-analysis
length(res$Meta)
#To plot an histogram of raw p-values
rawpval=2*(1-pnorm(abs(res$TestStatistic)))
hist(rawpval,nclass=100)

P-value combination for paired data

Description

Calculates differential expression p-values from paired data either from classical or moderated t-tests (Limma, SMVar) for each study and combines these p-values by the inverse normal method.

Usage

pvalcombination.paired(logratios, moderated = c("limma", "SMVar", "t")[1], BHth = 0.05)

Arguments

Value

List

Note

While the invisible object resulting from this function contains the values described previously, other quantities of interest are printed: DE,IDD,Loss,IDR,IRR. All these quantities are defined in function IDDIDR and in (Marot et al., 2009)

Author(s)

Guillemette Marot

References

Marot, G., Foulley, J.-L., Mayer, C.-D., Jaffrezic, F. (2009) Moderated effect size and p-value combinations for microarray meta-analyses. Bioinformatics. 25 (20): 2692-2699.

Examples

data(Singhdata)
#create artificially paired data:
artificialdata=lapply(Singhdata$esets,FUN=function(x) (x[,1:10]-x[,11:20]))
#Meta-analysis
res=pvalcombination.paired(artificialdata)
#Number of differentially expressed genes in the meta-analysis
length(res$Meta)
#To plot an histogram of raw p-values
rawpval=2*(1-pnorm(abs(res$TestStatistic)))
hist(rawpval,nclass=100)

Row t-tests

Description

Performs t-tests for unpaired data row by row.

Usage

row.ttest.stat(mat1, mat2)

Arguments

Details

This function is much faster than employing apply with FUN=t.test

Value

Vector with t-test statistics

Examples

## The function is currently defined as
function(mat1,mat2){ 
n1<-dim(mat1)[2]
n2<-dim(mat2)[2] 
n<-n1+n2 
m1<-rowMeans(mat1,na.rm=TRUE) 
m2<-rowMeans(mat2,na.rm=TRUE) 
v1<-rowVars(mat1,na.rm=TRUE) 
v2<-rowVars(mat2,na.rm=TRUE) 
vpool<-(n1-1)/(n-2)*v1 + (n2-1)/(n-2)*v2 
tstat<-sqrt(n1*n2/n)*(m2-m1)/sqrt(vpool) 
return(tstat)}

Row paired t-tests

Description

Performs t-tests for paired data row by row.

Usage

row.ttest.statp(mat)

Arguments

Details

This function is much faster than employing apply with FUN=t.test.

Value

Vector with t-test statistics.

Examples

## The function is currently defined as
function(mat){ 
m<-rowMeans(mat,na.rm=TRUE) 
sd<-rowSds(mat,na.rm=TRUE)  
tstat<-m/(sd*sqrt(1/dim(mat)[2])) 
return(tstat)}

Row variance of an array

Description

Calculates variances of each row of an array

Usage

rowVars(x, na.rm = TRUE)

Arguments

Details

This function is the same as applying apply with FUN=var but is a lot faster.

Value

A numeric or complex array of suitable size, or a vector if the result is one-dimensional. The dimnames (or names for a vector result) are taken from the original array.

Examples

## The function is currently defined as
function (x,na.rm = TRUE) 
{
    sqr = function(x) x * x
    n = rowSums(!is.na(x))
    n[n <= 1] = NA
    return(rowSums(sqr(x - rowMeans(x,na.rm = na.rm)), na.rm = na.rm)/(n - 1))
  }