Deseq: Multiple Conditions Testing

Question

[k.nirmalraman]

I am finding it a bit hard to comprehend a case where multiple conditions are to be compared and find myself confused with the multifactor part of the DESeq Vignette..

Basically I have four tissues (conditions) and would like to find DE genes for all pairwise comparisons across all conditions. Is the following pairwise comparison approach in DESeq right?

>design
condition    libType
 1            single-end
 1            single-end
 1            single-end
 1            single-end
 1            single-end
 1            single-end
 1            single-end
 1            single-end
 1            single-end
 2            single-end
 2            single-end
 2            single-end
 3            single-end
 3            single-end
 3            single-end
 4            single-end
 4            single-end
 4            single-end
 4            single-end
 4            single-end
 4            single-end


cds <- newCountDataSet( ceiling(RawCountData), conds )
cds <- estimateSizeFactors( cds )
sizeFactors( cds )
cds<-estimateDispersions(cds)

#contrasts for differential expression in DE
 res12<-nbinomTest(cds,1,2)
 res13<-nbinomTest(cds,1,3)
 res23<-nbinomTest(cds,2,3)
 res24<-nbinomTest(cds,2,4)
 res34<-nbinomTest(cds,3,4)

Any better ways of doing it? Also as an extension, I am not sure yet, if the condition 1 and 4 have different time points (which I presume to be, however, I am not interested in time point specificty), will this by any chance change the approach?

Thanks!

Answer 1

[venks]

Hi,

It looks fine. But I guess if you read in your conditions as vector/matrix that would make life easier. Btw I am wondering if you are doing it that way?

[k.nirmalraman]

Thanks!! That reassured... Yes, I am reading it as a phenoData Matrix and using it as vector

[venks]

Perfect! I have never done DESeq with pairwise comparison but I have used DEXSeq which is almost same as DESeq. I guess you are doing it right. Good luck.

Answer 2

[Toni]

Is there any way to test for differential expression of all four conditions together instead of pairwise?

[Diana]

It would be quite good to know that! I've been trying to find that out as well

Answer 3

[Charles Warden]

If you have multiple variables, you can use the GLM function in DESeq.

Linear regression can provide the analysis that you are looking for, so I would assume you could use the DESeq implementation to model a single factor with 4 levels. See the overview starting at page 10 for a description of the multi-factor analysis in DESeq:

http://bioconductor.org/packages/release/bioc/vignettes/DESeq/inst/doc/DESeq.pdf