Question about using limma for differential analysis

[bvk]

For the Differential analysis I'm using "limma" R package. I'm using RSEM log2 transformed counts data from XENA browser. A matrix "U" with 20,000 genes as rows and more than 300 samples as columns. Initially I have done some filtering steps like following:

y <- normalizeQuantiles(U)

keep <- rowSums(y > log2(11)) >= 14

table(keep)
keep
FALSE TRUE
3624 16906

y2 <- y[keep,]

table(G$MB)

MB1 MB2
286 80

design2 <- model.matrix(~ G$MB -1)
colnames(design2) <- c("MB1","MB2")

head(design2)
MB1 MB2
1 1 0
2 1 0
3 0 1
4 1 0
5 1 0
6 1 0

fit <- lmFit(y2,design2)
contrast.matrix <- makeContrasts(MB2-MB1, levels=design2)

contrast.matrix
Contrasts
Levels MB2 - MB1
MB1 -1
MB2 1

fitC <- contrasts.fit(fit,contrast.matrix)
fitC <- eBayes(fitC,robust=TRUE,trend=TRUE)
tab <- topTable(fitC,adjust="BH",n=Inf)

summary(decideTests(fitC))
MB2 - MB1
-1 1
0 16904
1 1

Does this mean there are two differentially expressed genes between the "MB2-MB1" comparison?

And when I did like follwoing:

sig.deg = subset(tab, abs(logFC)>0.5 & tab$P.Value < 0.05)

head(sig.deg)
logFC AveExpr t P.Value adj.P.Val B
PRR18 -0.7536122 5.163854 -3.396209 0.0007577785 0.3583424 -0.7996981
DACH2 0.6016372 0.845996 3.375142 0.0008163282 0.3583424 -0.8561809
HELLS 0.5986486 5.800163 3.353323 0.0008813889 0.3583424 -0.9143339
OSBPL6 0.6416259 5.887980 3.226050 0.0013674819 0.412833 -1.2465415
POLQ 0.5932946 5.515293 3.212282 0.0014328446 0.4131078 -1.2817619
MYH1 -0.6999817 1.053430 -3.211477 0.0014367504 0.4131078 -1.2838148

Here the genes are with adj.P.Val > 0.05. I don't understand which are differentially expressed genes. my statistical knowledge is pretty low. Can anyone tell me about this?

[neavemj]

You will want to use the "adj.P.Val" column. These are the p-values that have been adjusted for multiple testing - you have done a lot of comparisons, so this needs to be taken into account. For example:

sig.deg = subset(tab, abs(logFC)>0.5 & tab$adj.P.Value < 0.05)

Perhaps you indeed have very few differentially expressed genes, which could be for a variety of reasons. You might want to do an initial PCA ordination to check if your replicates are similar to each other. If there is as much variation within your replicates as there is across the treatment groups, then few DE genes will be detected.

Good luck!

Matt.