I have seen this question posed elsewhere, but wanted to double check for some of my colleagues who want to make sure. It is the following: when defining contrasts in Limma, how is the syntax to be interpreted? I've included some code. If clarification is required, I can provide that as needed.

#### edata is expression matrix of aptamer probes; design is the design matrix; Used a repeated measures design so estimated the patient-wise correlation in expression over visits where StudyID defines this pairing ; pheno_table details sample metadata 

fit<-lmFit(edata,design,
               block=pheno_table$StudyID,
               correlation=corfit$consensus)

cm <- makeContrasts(
    `Diabetpost.PUFA.Met-Diabetpre.PUFA.Met` = Diabetpost.PUFA.Met-Diabetpre.PUFA.Met,
    `Diabetpost.MUFA.Met-Diabetpre.MUFA.Met` = Diabetpost.MUFA.Met-Diabetpre.MUFA.Met,
        `Diabetpost.PUFA.NoMet-Diabetpre.PUFA.NoMet` = Diabetpost.PUFA.NoMet-Diabetpre.PUFA.NoMet,
    `Diabetpost.MUFA.NoMet-Diabetpre.MUFA.NoMet` = Diabetpost.MUFA.NoMet-Diabetpre.MUFA.NoMet,
     levels=design)

fit2 <- contrasts.fit(fit, cm)
fit2 <- eBayes(fit2,robust=TRUE)
T<-topTable(fit2,number=100000)

In plain English, are the above comparisons made "between post with respect to pre" or "between pre with respect to post"? That is, would a positive logFC indicate post > pre or pre > post? Again, I believe I know the answer, but I would like confirmation for my colleagues. Thanks.

It's easy to understand in terms of how you called makeContrasts. Take your first comparison:

    `Diabetpost.PUFA.Met-Diabetpre.PUFA.Met` = Diabetpost.PUFA.Met-Diabetpre.PUFA.Met,

The log-fold change will be computed as Diabetpost.PUFA.Met-Diabetpre.PUFA.Met, i.e., the difference between the two coefficients. So if the log-fold change is positive, you've basically got:

Diabetpost.PUFA.Met - Diabetpre.PUFA.Met > 0 # i.e.,
Diabetpost.PUFA.Met > Diabetpre.PUFA.Met

So a positive log-fold change would represent an increase in "Post" relative to "Pre". And so forth.

I have also wondered this question. How I came to my solution was to check a single gene that had a log-fold change and then go look at the CPM values for the same gene. It is very easy to tell by this method and there is basically no chance to misinterpret the answer.