How do I work out which condition the genes are down-regulated in?

Question

[Tota]

Hi all, I am using edgeR to find DEG. I have paired samples for cancerous tissue and adjacent non-cancerous tissue.

> type <- factor(c("benign", "tumour", "benign", "tumour", "benign", "tumour") 
> subject <- factor(c(1, 1, 2, 2, 3, 3)) 
> design <- model.matrix(~subject+type)
> keep <- filterByExpr(y)
> y <- y[keep, , keep.lib.sizes=FALSE]
> y <- calcNormFactors(y)
> fit <- glmQLFit(y,design)
> qlf <- glmQLFTest(fit)

Then I used glmQLFit and glmQLFTest and got the below result for one gene for example

> Gene                       logFC          logCPM      F           PValue      FDR 
> ENSG00000165186.12|PTCHD1 -4.570433395    5.571673968 88.74868282 2.53E-09    6.31E-06

What I'm trying to understand for the logFC in which condition is it down-regulated? Looking at the cpm table output, I can't see in which condition it is down-regulated (or up-regulate).

Thank you.

[rpolicastro]

Since you didn't include all of your code we can't see the exact contrast you did, but the wald test will by default compare each factor to the first factor level. For subject the first factor level is 1, so the contrasts would be 2 vs 1 and 3 vs 1, and for type the first factor level is benign so the contrasts would be tumor vs benign.

[Tota]

Perfect, thank you. Is it right to say the first factor level for type is benign because it's written first? Initially I thought it would be alphabetical. I'll add all my code.

[ATpoint]

By default the QLFTest tests the last column of the design because the preset is coef=ncol(glmfit$design). You have to change it is you want individual contrasts/coefs being tested.

[Gordon Smyth]

Just noting that edgeR doesn't use Wald tests. The above code will compute quasi-F-tests where the numerator is the likelihood ratio test statistic.

Answer 1

[Gordon Smyth]

The logFCs are for tumour vs benign. A negative logFC means that the gene is down-regulated in tumour compared to benign.

The output from topTags(qlf) tells you which contrast is being tested. The first line of output will say Coef: type:tumour, which is telling you that the contrast is for the tumour level of type vs the reference level.