Dear EDGER support team,

I have a single factor experiment with the 5 different samples having three replicates each.

group <- factor(c("a","a","a","b","b","b","c","c","c","d","d","d","e","e","e"))

Using a GLM based approach, and using TreatDGE function, I am trying to obtain the Differentially expressed genes. The following two kinds of comparisons are of interest to me:

1. (b vs a), (c vs a), (d vs a), (e vs a)

2. (c vs b), (e vs d)

For the 1st type of comparison, I was using this:

tr <- treatDGE(fit, coef=2, lfc=1) (b vs a)

tr <- treatDGE(fit, coef=3, lfc=1) (c vs a)

tr <- treatDGE(fit, coef=4, lfc=1) (d vs a)

tr <- treatDGE(fit, coef=5, lfc=1) (e vs a)

Is this correct?

How can I make use of the "contrast" argument of the TreatDGE function for these comparisons.

And how to do the second kind of comparisons? I read the TreatDGE function information provided, but not able to decide on how to use it correctly.

Thanks in advance for the help!

Candida

I'm assuming that your fit used the design matrix:

design <- model.matrix(~group)

To test (c vs b) you can use:

treatDGE(fit, contrast=c(0,-1,1,0,0), lfc=1)

To test (e vs d):

treatDGE(fit, contrast=c(0,0,0,-1,1), lfc=1)

Note that

treatDGE(fit, coef=2, lfc=1)

is just shorthand for:

treatDGE(fit, contrast=c(0,1,0,0,0), lfc=1)

In general, the contrast argument behaves the same for treatDGE() as it does for the glmLRT() function.

Dear Gordon,

Thanks a lot for your prompt reply,

I read in the edge R tutorial that "0+ in the model formula is an instruction not to include an intercept column and
instead to include a column for each group".

so I decided to fit my design matrix as: "design <- model.matrix(~0+group)"

x <- read.delim("known-miR-expression-profiling-max-read-counts.txt",row.names="Symbol")
group <- factor(c("a","a","a","b","b","b","c","c","c","d","d","d","e","e"))
y <- DGEList(counts=x,group=group)
data.frame(Sample=colnames(y),group)
y <- calcNormFactors(y)
norm.vals <- cpm(y)
design <- model.matrix(~0+group)
y <- estimateGLMCommonDisp(y, design, verbose=TRUE)
y <- estimateGLMTrendedDisp(y, design)
y <- estimateGLMTagwiseDisp(y, design)
fit <- glmFit(y, design)

for b vs a :

groupbvsgroupa <- makeContrasts(groupb-groupa, levels=design)
tr <- treatDGE(fit, contrast=groupbvsgroupa, lfc=1)

for c vs b :

groupcvsgroupb <- makeContrasts(groupc-groupb, levels=design)
tr <- treatDGE(fit, contrast=groupcvsgroupb, lfc=1)

Is this a correct way too?

Another doubt I had is that is one of the sample replicate is not good and I decide to remove it, so for 4 samples  (a-d)

I have 3 replicates and the 5th one (e) has two, can I still proceed as:

group <- factor(c("a","a","a","b","b","b","c","c","c","d","d","d","e","e"))

Thank you very much for your kind help!

Candida