[BioC] Using DESeq2: experimental design and extracting results

Michael Love michaelisaiahlove at gmail.com
Tue Jun 10 21:38:15 CEST 2014

hi Sridhar,

On Tue, Jun 10, 2014 at 3:05 PM, Sridhar A Malkaram
<smalkaram at wvstateu.edu> wrote:
> Hi,
> I have been a user of DESeq and recently DESeq2 for my research work.
> The latest DESeq2 seem to offer extensive differential testing options
> suitable for various experimental designs.
> Recently I wanted to use DESeq for a differential gene expression
> analysis between two plant genotypes across 4 different time points.
> I am basically a biologist and am finding hard to grasp the concepts of
> testing results. I'd be very grateful if you could help me understand
> some concepts (especially resultsNames) related to the DESeq2 package.
> My experimental design is as below
> design<- ~ genotype + time + genotype:time
> There are two levels in genotype and 4 levels in time.
> Basically I'd like to use binomLRT test to check if there is any
> difference in gene expression between the genotypes across the time points.
> dds<-DESeq(dds)  (dds is DESeq2 object  obtained from,
> dds<-DESeqDataSetFromMatrix(countData=counts, colData=coldata,
> design=design)
> and I am using the reduced model for the liklihood test

Here is where things are getting confused. You have already run
DESeq() using test="Wald". So it doesn't make sense at this point to
instead perform a likelihood ratio test. In our vignette we explain
this in the section on the LRT: "The likelihood ratio test can also be
specified using the test argument to DESeq, which substitutes
nbinomWaldTest with nbinomLRT."

> Is the model correct per my research question (is there a (time
> influenced) difference  between genotypes)?

Yes. If you want to find those genes which show a time influenced
difference between genotypes, this is simply:

dds <- DESeq(dds, test="LRT", reduced=genotype + time)
res <- results(dds)

You can then use heatmaps to inspect the patterns of gene expression
for the differentially expressed genes. Visualization with heatmaps
are covered in the vignette.

If you have other more specific questions about how to generate
results tables, I can answer them. With time series experiments, there
are many possible combinations to test, but rather than going through
all combinations, we recommend that users explore the results with


> Thanks,
> Sridhar Acharya
>         [[alternative HTML version deleted]]
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor

More information about the Bioconductor mailing list