Hi All,

I've gone through and done all my contrasts for a rather large DE experiment using edgeR. I have logFC, logCPM, LR, PValue, and FDR for each of these contrasts but I can't figure out how to add a column that shows up/down regulation for each of my DEGs. I've tried extracting this information using 

updown.12Cvs24C<-decideTestsDGE(lrt.12Cvs24C, p.value=0.05)

and then adding this column to the results tables using cbind, but I don't think the up/down values are correctly corresponding to the right genes because I have a list of the DEGs (filtered for an FDR < 0.05) that show SAME (0) up/down regulation. If the FDR of these genes is < 0.05 (like the 6 genes shown below), it means that there expression was either significantly higher or lower in one group vs the other, yet the UpDown column doesn't reflect this.

  geneID    logFC      logCPM       LR       PValue         FDR            UpDown
237542  -5.901809 2.871127 41.81711 1.002224e-10 5.653145e-07      0
1879515 -6.603091 5.203004 41.70325 1.062322e-10 5.653145e-07      0
1207284 -6.666918 4.879160 40.38631 2.083963e-10 7.393207e-07      0
131752  -5.096021 7.731907 37.80230 7.828952e-10 2.049564e-06     -1
29878   -6.932289 6.619459 37.25548 1.036232e-09 2.049564e-06      0
483711  -8.104056 7.199398 37.02868 1.164046e-09 2.049564e-06      0

Is there an easy way to extract this information and add it to the results table or do most people just infer it from the logFC values that are already present in the results table? FYI- not an R wiz so be gentle.

Bonus question (while i have your attention): Is there an easy way to isolate the DEGs that are similar between two groups (different contrasts). I know this can be done easily in Galaxy but I'd like to learn it in R. So far can't find a lot about this kind of action.

thanks,

carrie

Hi Carrie,

I think the order of genes in your results (topTags) table is different than the order in lrt.12Cvs24C.

You can try to use the topTags function with sort.by="none" and see if it does matches.

Succes!

Ben

As for your bonus question, this depends on how you define "similar". If you want to find the genes that are DE in both contrasts, the simplest approach is to just intersect the DE lists:

sig.1 <- decideTestsDGE(result.1, p.value=0.05)
sig.2 <- decideTestsDGE(result.2, p.value=0.05)
both <- sig.1 != 0 & sig.2 != 0

The both vector then indicates the genes that are significant in both contrasts at a FDR of 5%. If you want to be more statistically rigorous, you can do an intersection-union test before controlling the FDR:

both.pval <- pmax(result.1$table$PValue, result.2$table$PValue)
both.fdr <- p.adjust(both.pval, method="BH")
both <- both.fdr <= 0.05

However, if you want to find DE genes that have the same expression in two groups, this is much harder to do. This is because the majority of the testing machinery (here and elsewhere) is built to test for differences, not for similarities - absence of evidence is not evidence of absence, and so on. Trying to identify genes with similar expression in edgeR is a bit ad hoc, and just involves selecting those genes with large p-values for the contrast between the two relevant groups.