Hello people, i have a big microarray data set, in which i want to perform a paired t-test among the rows representing each gene, for two indices which consist of the two different groups(samples) to be compared. My samples are 34. Here is my code :

globalt.test <- function (ExpressionSet, ig1, ig2) {
  for( i in 1:nrow(ExpressionSet) ) {
    ig1 <- ExpressionSet[i, c(1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33)]
    ig2 <- ExpressionSet[i, c(2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34)]
    tt <- t.test(exprs(ig1), exprs(ig2), paired=TRUE, var.equal=FALSE)
  }
  u <- return(tt[i])
}

But the function makes a lot of time(there are 54675 features) and then returns:

$NA
NULL

Any ideas or suggestions ??

Yes. Use limma instead.

library(limma)

trt <- factor(rep(1:2, 17))

pairs <- factor(rep(1:17, each = 2))

design <- model.matrix(~trt+pairs)

fit <- lmFit(ExpressionSet, design)

fit2 <- eBayes(fit)

topTable(fit2,2)

and see the limma User's Guide: http://bioconductor.org/packages/release/bioc/vignettes/limma/inst/doc/usersguide.pdf

Dear James W. MacDonald, 

i would also like to ask you about the above aproach you mentioned :

if based on the above phenotype i use limma :

condition <- factor(ExpressionSet$tissue)
pairs <- factor(rep(1:17, each = 2))
design <- model.matrix(~condition + pairs)
fit <- lmFit(ExpressionSet, design)
ebayes <- eBayes(fit)

because im very new in Bioconductor and statistical interference, if i change the design object: design2 <- model.matrix(~pairs + condition), what is the main difference with the first design object??

I didnt implemented like this in limma. In other words, i have used limma to compare two groups for one variable from my pData object. 

pData(agcrma)
                                    meta factor            tissue
St_1_WL57.CEL             0                              0
St_2_WL57.CEL             0                              1
St_N_EC59.CEL             0                              0
St_T_EC59.CEL             0                              1
St_N_EJ58.CEL             0                               0
St_T_EJ58.CEL             0                               1
St_N_FD58.CEL             0                               0
St_T_FD58.CEL             0                               1
St_N_FG39.CEL             1                               0
St_T_FG39.CEL             1                               1
St_1_GG52.CEL             1                               0
St_2_GG52.CEL             1                               1
St_N_HJ33.CEL             0                                0
St_T_HJ33.CEL             0                                1
St_N_HK57.CEL             0                                0
St_T_HK57.CEL             0                                1
St_N_KA49.CEL             0                                0
St_T_KA49.CEL             0                                1
St_N_KW40.CEL             0                               0
St_T_KW40.CEL             0                               1
St_1_LH53.CEL             0                                 0
St_2_LH53.CEL             0                                 1
St_N_LW46.CEL             0                                0
St_T_LW46.CEL             0                                1
St_N_NH45.CEL             0                                0
St_T_NH45.CEL             0                                1
St_N_PH42.CEL             0                                0
St_T_PH42.CEL             0                                1
St_1_SSch60.CEL           0                               0
St_2_SSch60.CEL           0                               1
St_N_WH37.CEL             0                               0
St_T_WH37.CEL             0                               1
St_N_WM60.CEL             0                              0
St_T_WM60.CEL             0                              1

design <- model.matrix(~factor(agcrma$tissue)) 

 fit <- lmFit(afilter, design)

ebayes <- eBayes(fit, trend=TRUE)

names <- unlist(mget(featureNames(afilter), env=hgu133plus2SYMBOL))

tab <- topTable(ebayes,coef=2, adjust.method="fdr", genelist=names)

I have read the vignette and other tutorials for limma but i didnt know that paired samples could be implemented. My approach is very different from the above suggestion ?
 

Thank you for your important recommendation !!! i didnt mentioned this in my previous answer, but when i get the data i had totally 17 patients, each with a control sample and a cancer sample respectively . So in your opinion i should use the limma aproach but with paired samples, instead of just define two categories, with control samples and cancer samples respectively as my code ? In other words, It is more appropriate to consider the association of the two samples in each patient for performing my statistical analysis in order to get more powerful results and also biologically significant ?