Hello,
I have an experiment with six groups: WT and KO cells that are untreated, treated with IFN A or IFN B.
Here are some of the possible contrasts.
cm <- makeContrasts( # Treatment effects A.WT = WT.A - WT.Un, B.WT = WT.B - WT.Un, A.KO = KO.A - KO.Un, B.KO = KO.B - KO.Un, # Differences between treatments within strain A.vs.B.WT = WT.A - WT.B, A.vs.B.KO = KO.A - KO.B, # Basal difference between strains KO.vs.WT = KO.Un - WT.Un, # Treatment differences between strains A.KO.vs.WT = (KO.A - KO.Un) - (WT.A - WT.Un), B.KO.vs.WT = (KO.B - KO.Un) - (WT.B - WT.Un), levels = design ) fit2 <- contrasts.fit(fit, cm) fit2 <- treat(fit2, lfc = log2(1.2)) results <- decideTests(fit2)
One of the main questions of interest is what genes are up-regulated specifically by the treatments. e.g. in WT cells, genes that are significantly induced by A compared to unstimulated and to B, but not induced by B. The common specific genes in WT and KO cells are also of interest.
A.WT.specific <- results[,"A.WT"] == 1 & results[,"A.vs.B.WT"] == 1 & results[,"B.WT"] != 1 # Similarly for A.KO.specific, B.WT.specific and B.KO.specific A.specific <- A.WT.specific & A.KO.specific # Similarly for B.specific
I think this is a situation where using decideTests
with method="global"
would be appropriate, however I want to avoid including unnecessary contrasts.
I was wondering whether anyone could offer any advice about whether the contrast matrix above would be suitable to use with the "global"
method or whether I would need to define smaller contrast matrices, such as those shown below, and run contrasts.fit
, treat
and decideTests
again as appropriate.
Thanks very much.
# Keep both WT and KO together cm.specific <- makeContrasts( # Treatment effects A.WT = WT.A - WT.Un, B.WT = WT.B - WT.Un, A.KO = KO.A - KO.Un, B.KO = KO.B - KO.Un, # Differences between treatments within strain A.vs.B.WT = WT.A - WT.B, A.vs.B.KO = KO.A - KO.B, levels = design ) # Split even further into WT and KO contrast matrices cm.wt.specific <- makeContrasts( # Treatment effects A.WT = WT.A - WT.Un, B.WT = WT.B - WT.Un, # Differences between treatments within strain A.vs.B.WT = WT.A - WT.B, levels = design ) # cm.ko.specific similarly
Thanks Aaron. I was leaning towards that particular choice, but it was very helpful to get another opinion. Yes, I shall just subset
fit2
for thedecideTests
and then go from there.I had a suspicion that perhaps this was not entirely statistically rigorous, particularly the part about which genes are not significantly different, but it was best way that I could think of. It seems to get me a few candidate genes and I suppose it is better than nothing.
Thanks again.