Hi, I am trying to normalize RNA-Seq using DESeq2. I went on their website and they have three different transformations out of which I tried vst (variance stabilization transformation) and rlog. The thing is that if I have only sample per condition so vst didn't work in those cases but rlog did. Could someone explain me how different they really are? They show the graph on the guide page, and I can see that the lower values in rlog are different than the vst, but the higher count values are not much different. I want to use the normalized data to do WGCNA, where they mention using vst data. I want to know if rlog is similar scale?
Thanks!
1 answer
I think rnaseqGene has a very nice DESeq2-centric tutorial with a helpful visualization and explanation.
At least for me, this difference wasn't as clear from reading the papers, but the VST normalized counts in that figure look more like thresholded counts.
Log in to answer this question.
-
Weird correlation between RIP-seq replicates
written by Hiba •
-
Inquiry about deseq2 transformation
written by Chen •
-
Using DESeq2 normalized counts or VST transformed counts for model building
written by alexandriashumway •
-
What is the importance of the rlog function (DESeq2) for downstream analysis?
written by Aspire
-
Transforming RNAseq data with widely varying library sizes
written by adampennycuick
-
what is inverse normalized RNA-seq expression and how to do that?
written by piyushjo
-
Check variance stabilization transformation ?
written by debitboro
-
DESeq2 VST question
written by MarcHiggins •
-
variance stabilized or log transformed data for DE analysis
written by tonja.r
-
DeSeq2 data transformation for visualization
written by sunil.mangalam •
Did you read this one already: Question: Is there a conceptual difference between vst/rlog transforms and lfcShrink??
In my understanding, vst takes two things into account while normalization, one is size factor(depth normalization) and another is mean dispersion across the sample. while rld only accounts for size factor.
rlog tends to work well on small datasets (n < 30)