Dear all,

I would like to use cameraPR() to test for the enrichment of some gene-sets (e.g. KEGG, GO, Reactome...) without specifying any direction as some genes can be up and others down within the same gene-set. According to the CAMERA's paper (http://nar.oxfordjournals.org/content/early/2012/05/24/nar.gks461.full) we can test for "non-directional" gene sets using the absolute log(FC) as statistic and I was wondering whether I could use this (or absolute t-statistic) with cameraPR(). For instance with this code:

y <- matrix(rnorm(1000*6),1000,6)
design <- cbind(Intercept=1,Group=c(0,0,0,1,1,1))
# First set of 20 genes are genuinely differentially expressed (up)
index1 <- 1:20
y[index1,4:6] <- y[index1,4:6]+1
# Second set of 20 genes are not DE
index2 <- 21:40
# Third set of 20 genes are down
index3 <- 41:60
y[index3,4:6] <- y[index3,4:6]-1
# fit model
fit <- eBayes(lmFit(y, design))
cameraPR(abs(fit$t[,2]), list(set1=index1, set2=index2, set3=index3), 
         inter.gene.cor=0.01, use.ranks=FALSE)
#     NGenes Direction      PValue         FDR
#set1     20        Up 0.002350959 0.007052876
#set2     20      Down 0.719682160 0.719682160
#set3     20        Up 0.025908100 0.038862150

In the output set3 appears to be "Up" as its genes' t-statistics are at the top of the list. Would this approach be correct?

Thanks for your help, best regards,

Hugo

Yes, you can still do non-directional tests using cameraPR(), but it's not quite as easy as it was using the original camera() function as it was in 2012 because we no longer output the one-sided p-values for each gene set. We now output the two-sided p-values only, in order to simplify the output.

If you input absolute t-statistics (recommended) or absolute logFC as the statistic, then you should set use.ranks=TRUE in the call to cameraPR(). You also need to convert the two-sided p-values back to upper-tail p-values by:

PValue <- out$PValue / 2
Dn <- out$Direction == "Down"
PValue[Dn] <- 1 - PValue[Dn]
FDR <- p.adjust(PValue, method="BH")

where out is the output from cameraPR().

In the next version of limma (3.64.0), I will add in a directional argument to camera() and cameraPR() to make non-directional tests easier by setting directional=FALSE.