[BioC] opposite signs in limma

[Gordon Smyth]

At 05:16 AM 3/08/2005, Jihoon Kim wrote:
>Dear Dr.Smyth,
>
>I'm struggling with limma to apply it to two-group microarray data.
>I thought ebayes() in limma shrinks the variance but not the "sign"
>of the original t-statistic.

It does.

>  However, example code resulted in opposite
>signs of the moderated T statistic and  the mean difference
>in the gene 1(or the row 1).
>
>I tried to follow example code in limma.
>What seems to be problem here?

Your meanDiff() is not equivalent to fitting a linear model. All your 
linear model does here is to compute group means. Suggest you talk to a 
statistician in your dept.

Gordon

>Thank you in advance.
>
>
>Best,
>
>
>Jihoon Kim
>
>
>--------------------------------------------------------------
>
>
>library(limma)
>
>
>exprVec <- c(
>    1.5522,   1.6881,  1.0798,     1.3877,   0.8566,   1.4138,
>    1.5729 ,  1.3380,   2.2301,    1.8566,   2.0400 ,  1.6191,
>  -1.6955,  -1.2777, -1.4938,  -1.7326, -1.5530,  -1.7169,
>  -0.1812,    0.0399,   0.0413,  -0.3797, -0.4582,  -0.6490,
>  -0.3439,  -0.2067, -0.0792,    0.0549, -0.1200,   0.4584,
>  -0.2357,  -0.5873, -0.0446,  -0.1868,   0.2710,  -0.2091,
>  -0.5146,  -0.1384, -0.1525 ,   0.0089,   0.5896,   0.2688,
>   0.2017,     0.1631,  -0.1724, -0.5886, -0.1346, -0.3309 )
>Mat <- matrix( exprVec,  nrow=4,  ncol=12, byrow=TRUE)
>design.eb <- cbind(   time1=c( rep(1, 6), rep(0, 6)),
>                      time2=c( rep(0, 6), rep(1, 6))     )
>fit <- lm.series(Mat, design=design.eb)
>eb <- ebayes(fit)
>modT <- eb$t[,1]
>
>meanDiff <- rep(0, 4)
>for(i in 1:3) {
>    meanDiff[i] <- mean( Mat[i, 1:6] )  - mean( Mat[i, 7:12] )
>}
>
>
>cbind(modT, meanDiff)
>
>plot(1:12, Mat[1,], type="b")