different results with normal fit and contrast matrix fit

Question

[Ice P]

I have a data set with loge base expression values of 90 normal and 10 tumor sample.

mydesign<- cbind(Tumor=c(rep(1,90),rep(0,10)),Control=c(rep(0,90),rep(1,10)))

Now, I proceeded with two way:

fit <- lmFit(data, design=mydesign)
fit <- eBayes(fit)
topTable(fit)

numGenes <- nrow(data)
case1.list <- topTable(fit,number=numGenes)

or

cont.matrix <- makeContrasts(TumorvsControl=Tumor-Control, levels=mydesign)
fit2 <- lmFit(data, design=mydesign)
fit2 <- contrasts.fit(fit2, cont.matrix)
fit2 <- eBayes(fit2)

case2.list <- topTable(fit2,number=numGenes)

However, when I searched for a particular gene, the results are quite contrasting. e.g.

case1.list["ENSG1",]
                  Tumor  Control AveExpr        F      P.Value    adj.P.Val
ENSG1 1.53268 5.134342 1.87246 1802.497 1.115791e-83 4.074213e-83

case2.list ["ENSG1",]
                    logFC AveExpr         t      P.Value    adj.P.Val        B
ENSG1 -3.601662 1.87246 -29.42565 2.183333e-53 3.353327e-49 110.5354

Could anyone please advice that with given tumor and control estimation, how log FC were obtained?

[Antonio R. Franco]

"contrast" in limma means something similar to "What conditions I want to compare".

Answer 1

[Devon Ryan]

This ends up not being a limma question, but more of a question of what a model matrix is and how it's used.

Case 1 is asking, "Is the signal in the 'Tumour' group greater than 0?" That is unlikely to be a biologically interesting question to ask.

Case 2 is asking, "Is there a difference between 'Tumour' and 'Control'?" That is a much more biologically interesting question to ask.

I would strongly discourage you from manually making model matrices until you're more familiar with them (and even then it's rarely more convenient to do so).

[russhh]

This should be the accepted answer.

Answer 2

[Kamil]

Please read the documentation for limma. Start with Chapter 9 on page 40.