Hello,

I'd like to get some help on my RNAseq data analysis with DESeq2. The experiment is balanced design with multiple factors with biological/technical replicates.

> coldata
          group variety treatment
BeC-1     T      Be   control
BeC-2     T      Be   control
BeT-1     T      Be   treated
BeT-2     T      Be   treated
LoC-1     S      Lo   control
LoC-2     S      Lo   control
LoT-1     S      Lo   treated
LoT-2     S      Lo   treated
NiC-1     T      Ni   control
NiC-2     T      Ni   control
NiT-1     T      Ni   treated
NiT-2     T      Ni   treated
SaC-1     T      Sa   control
SaC-2     T      Sa   control
SaT-1     T      Sa   treated
SaT-2     T      Sa   treated
SoC-1     S      So   control
SoC-2     S      So   control
SoT-1     S      So   treated
SoT-2     S      So   treated
ViC-1     S      Vi   control
ViC-2     S      Vi   control
ViT-1     S      Vi   treated
ViT-2     S      Vi   treated

To eliminate the "model matrix is not full rank" as suggested in the manual, I used:

> coldata$ind.n <- factor(rep(rep(1:2, each=2), 2)
> model.matrix(~ group + group:ind.n + group:variety, coldata)
> dds <- DESeq(ddsHTSeq)

Now I want to test the significance for factors:

1) between groups:  T vs S;

2) among varieties,
3) between control vs treated for each variety: i.e. what are the genes differentially expressed for each variety Be, Lo, Ni, Sa, So and Vi.

I have tried 2) as following:

>results(dds, contrast=c("treatment", "treated", "control"))
>results(dds, contrast=c("group", "T", "S"))

But I am not sure for each individual variety, sth like this?

>results(dds, contrast=c("group$T.Be", "treated", "control"))
>results(dds, contrast=c("group$T.Lo", "treated", "control"))

Thanks a lot!

Can you explain what are the groups S and T biologically? Even vague descriptions would help.

And what are the varieties? Are these like species?

Thanks Michael!
My materials are plant/crops. S is susceptible vs T tolerant to stress. These are the 2 groups in trait/phenotype. Varieties can be called "genotypes" here, and they are different lines of the same plant/crop.

1) I am not sure my way to fix the "model matrix is not full rank" error is correct or not.

2) What is the syntax for each variety to test the significance between control and treated no matter it is in T or S group (i.e. after subtraction of the difference between S and T).   I was comparing this with the dissection of variance in ANOVA analysis with multiple factors design, which may not be appropriate. 

Thanks again.

 

Thanks!

Several points are in my mind:

1) Test the treatment vs control effect across T and S;

2) Test the treatment vs control effect within T and S, respectively;

3) Test the treatment vs control effect across all varieties (this should NOT be surprising if any significance turns out);

4) Test the treatment vs control effect within each variety, so that all the varieties can be aligned side-by-side, to see the bias of group selection.

What I did first is like you suggested "test treatment vs control within a variety". My concern is the normalization with DESeq, as the total reads count each sample may be different, but within each sample they may be similar, or vice versa. Putting all the sample in one set is for normalization, at least it is what I expect.

The major purpose is to test the difference between treatment and control for each variety across the two groups (T & S). "...... there are some conceptual issues with testing T vs S, because you also have the different varieties, and want to make inference on these as well. It makes it not so clear what exactly you want to test." ------the reason for different varieties is to have biological replicates.  It would not be surprising to see many genes are differentially expressed between treatment vs control no matter it is in T group or in S group. One big question behind: Is the difference due to the group (T) other than the treatment? This selection of group explains why "The varieties are not exactly random samples within the groups. These are a specifically chosen set of varieties."

So, a second step is to test the difference between treatment and control in each group (T, S separately). I was thinking integrating the group as the first factor may eliminate the bias of group selection.

Thank you for this discussion, and I appreciate any clarification for what I may have misunderstood.