test.LC: Test linear combinations of parameters of a linear model
In EMA: Easy Microarray Data Analysis

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

This function performs either a global significance F-test of one or several linear combinations of the parameters (contrast) of a (generalized) linear model, or tests for the significance of each contrast simultaneously, with or without multiple test correction.

1	test.LC(C, X, Y, global=FALSE, cor.multtest=TRUE,typeFDR="FDR-BH")

`C`	Vector or matrix of the linear combinations of factors we want to test.
`X`	Design matrix of the (generalized) linear model adjusted to the data.
`Y`	Vector or response matrix (e.g. a gene expression matrix) to which the (generalized) linear model is applied. If a matrix is specified the test(s) will be done on each column of the matrix.
`global`	logical indicating if one want to perform a F test of the global hypothesis H_0=0, or to test each contrast simultaneously
`cor.multtest`	logical.If global=FALSE indicates if one want to apply a multiple test correction for the computation of the p-values
`typeFDR`	If global=FALSE and cor.multtest=TRUE this argument is passed to the function multiple.correction(), it specifies which correction method to apply. See '?multiple.correction' for more details.

The design matrix X can be extracted from a (generalized) linear model with the function model.matrix. If Y is a response matrix each column shall be a gene/individual response on which the linear combinations will be tested. If one wish to test several linear combinations at the same time, C must be a matrix p*n with n the number of columns in the design matrix X (i.e. the number of parameters in the (generalized) linear model) and p the number of linear combinations, so that each row stands for one linear combination. Else C shall be a vector of length n. In case where both C and Y are matrices, each linear combination of the parameters (i.e. each row of C) will be tested for nullity on each gene/individual (i.e. each column of Y).

A list of matrices :

`Estimate`	A matrix containing the estimated values: F[i,j] is the estimated value of the linear combination i (on gene j).
`F`	A matrix containing the F values: F[i,j] is the F value of the test of linear combination i on gene j.
`pvalue`	A matrix containing the p-values: pvalue[i,j] is the p-value of the test of linear combination i on gene j.
`Y.pred`	A matrix containing the predicted response vectors
`resid`	A matrix containing the residuals of the linear model
`sigma2`	A matrix with one row containing the residual variance for each gene
`theta`	A matrix containing the estimates of the effects of the original model. These estimates are the ones also obtained with summary.lm()

Nicolas Servant, Eleonore Gravier, Pierre Gestraud, Cecile Laurent, Caroline Paccard, Anne Biton, Jonas Mandel, Bernard Asselain, Emmanuel Barillot, Philippe Hupe

model.matrix,lm

data(marty)

##Class label 0/1
marty.type.num <- ifelse(marty.type.cl=="Her2+",0,1)

#Annotation of the grade of tumor
grade=factor(sample(c(1:3),23,rep=TRUE),labels=c("I","II","III"))
typeXfactor=paste(marty.type.num,grade,sep=":")

##dataset fro example
normData<-marty[1:100,]

rt<-runTtest(normData, labels=marty.type.num, plot=FALSE)

normData.DE<-normData[which(rt["AdjpValue"]<0.05),]

marty.lm=lm(t(normData.DE)~as.factor(typeXfactor))
X=model.matrix(marty.lm)

#We want to test Basal vs Her2+ within each grade 
LC<-matrix(c(0,0,0,-1,0,0,0,1,0,0,-1,0,0,0,1,0,0,-1),ncol=6,byrow=TRUE)
#We also want to test grade II vs grade III 
LC2=c(0,1,-1,0,1,-1)
LC=rbind(LC,LC2)
row.names(LC)=c("B:I-H:I","B:II-H:II","B:III-H:III","II-III")
marty.LC=test.LC(LC,X,t(normData.DE))
marty.LC$pvalue

#List of the probesets differentially expressed for each of the four tests :
ll=list()
for(i in 1:nrow(marty.LC$pvalue)){
	ll[[i]]=as.matrix(marty.LC$pvalue[i,which(marty.LC$pvalue[i,]<0.05)])
	rownames(ll[[i]])=names(which(marty.LC$pvalue[i,]<0.05))
}
names(ll)=rownames(marty.LC$pvalue)
print(ll)