Hi all, 

I was just wondering if somebody could help me regarding normalized count after DESeq.

I carry out a DESeq on a time course experiment in order to then make a count matrix of significantly expressed genes to look for patterns in gene expression across different samples. However, I have noticed that the count matrix I am making post DESeq on significantly expressed genes are actually exactly the same as the initial raw counts table from which DESeq is working, so of course, am concerned the counts cannot be used as they have not been normalized. 

The code I am using is as follows. 

dds<-DESeq(dds reduced = ~Organ, test = "LRT")

and then:

res<-results(dds)
padj.cutoff<-0.05
sig_res <- subset(res, padj < padj.cutoff)
sigLRT_genes<- rownames(sig_res)
length(sig_res)... 2015 genes

orderedres <- sig_res[order(sig_res$padj), ]
clustering_sig_genes <- data.frame(orderedres[1:2015,])
rldres<-assay(dds)
cluster_rlog <- rldres[rownames(clustering_sig_genes),]
cluster_rlog... Count matrix

But the count matrix of cluster_rlog is identical to raw counts matrix. I therefore do not think that further downstream analysis is reliable. Am I do something wrong?

Any help would be much appreciated! Many thanks!

hi,

A quick answer to this one:

"However, I have noticed that the count matrix I am making post DESeq on significantly expressed genes are actually exactly the same as the initial raw counts table..."

DESeq2 models the raw counts, while using size factors or normalization factors to account for differences. This is also the approach of other statistical methods on Bioconductor like edgeR, EBSeq, etc. See the DESeq2 paper for more details on the methods, or the "Theory" section of the vignette.