Entering edit mode
Dear TS,
You are wanting to select genes which satisfy several conditions.
There is no way to do this with a single contrast. You need to
compute all three contrasts and then select genes:
cm = makeContrasts(ON=ON-WT, KO1=KO1-WT, KO2=KO2-WT, levels=design)
fit <- lmFit(data, design)
fit2 <- contrasts.fit(fit, cm)
fit2 <- eBayes(fit2)
results <- decideTests(fit2)
selectedgenes <- results[,"ON"]*results[,"KO1"]== -1 &
results[,"ON"]*results[,"KO2"]== -1
sum(selectedgenes)
write.fit(fit2[selectedgenes,], file="myresults.txt")
Unfortunately I don't know of any good introductions to linear models
which are suitable for a biology audience.
Best wishes
Gordon
>Date: Thu, 30 Nov 2006 21:23:15 -0500
>From: "T S" <tsm8015 at="" gmail.com="">
>Subject: [BioC] LIMMA makeContrasts question
>To: bioconductor at stat.math.ethz.ch
>Message-ID:
> <4a912b5b0611301823n5be74787o2e249ec9657041b6 at
mail.gmail.com>
>Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
>Hi,
>
>I am a complete beginner to Linear Models and using the LIMMA
package.
>First I have a general question as to what are good intro texts
>explaining Linear Models in general.
>
>I also have the following specific question. I have 4 samples: wild
>type (WT), a sample were I turn on a pathway (ON) and two where I
>knock down the same pathway in two different ways (KO1, KO2). I want
>to find genes that are significantly changed with the following
>property:
>
> sign(log(ON/WT)) == -sign(log(KO1/WT)) and -sign(log(KO2/WT))
>
>would the following design and contrast matrix be correct
>
>type = is a factor with levels WT, ON, KO1, KO2
>design = model.matrix(~0 + type)
>cm = makeContrasts(eff = (ON - WT) - (KO1 - WT) - (KO2 - WT),
levels=design)
>
>Any help or guidence would be greatly appreciated. Also pointers to
>suitable documentation explain such matters would be great.
>
>Thanks in advance
