Hi all, I've been using cuffdiff for a long time but recently I've given limma (voom) a try for RNA-seq data and compared the results.
The experiment is quite simple, a triplicate in a treatment vs control experiment (60M read tags for each sample). Cuffdiff finds almost 300 DGE, where limma finds almost 900. Almost 100 are in common between the two.
I have qPCR control for 3 genes and, according to the fold change, both the approaches are concordant with qPCR data. I understand that this is not an extensive test, but it is what can ben typically done within a lab (I mean, check few genes by qPCR). As a matter of facts, the FPKM quantification and the CPM quantification (cuffdiff vs limma) are quite consistent, the statistical test is making all the difference.
I've tried some pathway analysis to see if the two approaches are similar from the ontologic point of view... well, this is not really the case (there are some overlapping pathways, of course, but all in all the experiments are saying something different)
I see that others have tried such comparisons, with different results, now I realise I'm stuck in choosing the proper analysis approach... any hint?
d
The experiment is quite simple, a triplicate in a treatment vs control experiment (60M read tags for each sample). Cuffdiff finds almost 300 DGE, where limma finds almost 900. Almost 100 are in common between the two.
I have qPCR control for 3 genes and, according to the fold change, both the approaches are concordant with qPCR data. I understand that this is not an extensive test, but it is what can ben typically done within a lab (I mean, check few genes by qPCR). As a matter of facts, the FPKM quantification and the CPM quantification (cuffdiff vs limma) are quite consistent, the statistical test is making all the difference.
I've tried some pathway analysis to see if the two approaches are similar from the ontologic point of view... well, this is not really the case (there are some overlapping pathways, of course, but all in all the experiments are saying something different)
I see that others have tried such comparisons, with different results, now I realise I'm stuck in choosing the proper analysis approach... any hint?
d
Comment