**10**wrote:

Hi Michael,

I’ve tried to apply DESeq in two sample comparison, and it works well. It is really great for the NGS data analysis.

Now I tried to analysis multiple groups. Though I checked the section about glm model and refer to limma package and look for some answers in BioC posts, I was still stuck and confused. Will you help me to clarify it?

The first question:

For example, if the experiment contain two genotypes(”W“ and ”M“), and each with a factor Treatment (“U” and ”T” stand for “untreated“ and ”treated“ respectively). So I got four groups (“WU”,”WT”,”MU”,”MT”).

Now I want to find out genes response to treatment independent of genotypes, so I should compared WT vs WU, and get a differential expression gene set designed as A, while MT vs MU got a list B, and then find out the intersection of A and B(A∩B)?

If I want to find out genes that differentially response to treatment between genotype, i.e. the differences between set A and Set B as (A∪B-A∩B)?

Are those cases related to the formulas as follows:

fit0=fitNbinomGLMs(countdata,count~genotype+treatment)

fit1=fitNbinomGLMs(countdata,count~genotype+treatment+genotype:treatment

How should I design my formulas for that two cases (in DESeq or DESeq2)? And how to extract the results?

Some guys told me that they would like to design a group with four levels (“WU”,”WT”,”MU”,”MT”).

And code like:

fit0=fitNbinomGLMs (countdata, count~1)

fit1=fitNbinomGLMs (countdata,count~group)

And make pair contrast, and then combine results. Will it make the same thing?

The second question:

About the heat map. I read through the “Differential expression of RNA-Seq data at the gene level -the DESeq package”, and noticed that the heat map can be generated with raw counts after some kind of transform like “Variance stabilizing transformation” (DESeq2 might have different kind of transform). I may misunderstood, but the data for transform has not yet been tested? So whether it is significant is not yet supported by the statistic?

So can we just used the log2foldchange values?

fit0=fitNbinomGLMs (countdata, count~1)

fit1=fitNbinomGLMs (countdata,count~group)

pvalsGLM = nbinomGLMTest( fit1, fit0 )

padjGLM = p.adjust( pvalsGLM, method="BH" )

fit1.padj<-rbind (fit1, padjGLM)

After the filtration (padjGLM<0.1, make sure I did not misunderstand it, this value means, at least one group showed a significance with control?)

Use the log fold2change values to make a heatmap?

These questions trouble me a lot last week. I will appreciate your any kind help. Thank you.

Best wishes!

Chengzhi Liu