Hi, 

I am currently using DESeq (1.18.0) for an RNASeq analysis, in the

tutorial script, should the "estimateDispersions" function be taking normalized counts or non-normalized counts as input? Is it already using normalized counts without adding “/normalized=TRUE/” in the example script?

raw.counts<-newCountDataSetFromHTSeqCount(sampleTable=sampleTable, directory="./data/htseq")

cds<-estimateSizeFactors(raw.counts)

cds<-estimateDispersions(cds)

difftest<- nbinomTest(cds, "A", "B”)

normalized.counts<-counts(cds, normalized=TRUE)

Thank you very much!

Looks like you are doing everything correctly (syntax wise, anyway). If this is a new analysis (as opposed to trying to reproduce something you've seen or done a while ago) you are highly encouraged to upgrade to the latest DESeq2 to do your analysis, however.

Your first line is correct, but confusingly named. The following code is the same, but clearer:

cds <- newCountDataSetFromHTSeqCount(sampleTable=sampleTable, directory="./data/htseq")
cds <- estimateSizeFactors(cds)
cds <- estimateDispersions(cds)
difftest <- nbinomTest(cds, "A", "B”)
normalized.counts <- counts(cds, normalized=TRUE)

The first three lines all return a CountDataSet object. The first line creates one and the second and third line take such an object, add the resutls of their computational step to the information stored in the object, and return it.

So, estimateDispersion does not take a table of raw or normalized counts. It takes an object, whcih contains counts, and normalization factors, and other stuff, and find in it what all that it needs.

All this applies to DESeq2 as well, and as Steve already pointed out, we strongly recommend that people use DESeq2 and not DESeq for all new work.

Hi Steve and Simon,

Thank you for the advices! I will switch to DESeq2.

Yilong