Extracting within subject effect in a between and within subject model with DESeq2
1
0
Entering edit mode
i.sudbery ▴ 40
@isudbery-8266
Last seen 7 weeks ago
European Union

I am analysing an experiment where three mice of one genotype and 5 of another have been treated with a drug, and samples taken before and after treatment. This is thus a "between and within subject" design, with the design ~genotype + genotype:mouse + genotype:treatment i think.

We wish to ask: What is the effect of the transgene? What is the effect of the treatment? Is the effect of the treatment different between the two genotypes?

The PCA suggests PC1 is the identity of the mouse the sample came from, PC2 is the effect of the treatment. None of the first 5 PCs correspond to genotype.

My coldata (design) has columns sample_name, mouse, genotype, treatment.

I have created the model matrix as follows:

design$mouse <- factor(design$mouse)
design$mouse.n <- factor(c(rep(c(1,2,3), times=2), rep(c(1,2,3,4,5), times = 2)))
design$genotype <- relevel(design$genotype, "WT")
design$treatment <- relevel(design$treatment, "minus")

mm <- model.matrix( ~ genotype + genotype:mouse.n + genotype:treatment, data = design)

# Deal with the fact that there is no mouse 4 or 5 in the transgenic condition.
all_zero <- apply(mm, 2, function(x) all(x==0))
mm <- mm[, !all_zero]

I then executed the DESeq analysis so:

dds <- DESeqDataSetFromTximport(txi, colData=design, design=~genotype + treatment)
design(dds) <- ~ genotype + genotype:mouse + genotype:treatment
dds_processed <- DESeq(dds, full = mm)

Extracting the effects of genotype is easy(?)

genotype <- results(dds_processed, name = "genotypeTg")

And the interaction term:

diff_in_diffs <- results(dds_processed, contrast = list("genotypeTg:treatmentplus", "genotypeWT:treatmentplus"))

But what about the effect of treatment, accounting for per mouse and per genotype variance?

I think that worked out that it should be:

treatment_wt <- results(dds_processed, name =  "genotypeTg:treatmentplus")
treatment_tg <- results(dds_processed, name = "genotypeWT:treatmentplus")
treatment_all <- results(dds_processed, contrast = list(c("genotypeTg:treatmentplus", "genotypeWT:treatmentplus")), listValues = c(0.5, -1))

Is this right? Or the best way to do it? Since I'm looking for effects that don't (necessarily) differ between genotypes, would I be better re-analysing without the genotype in the design?

dds_treatment <- DESeqDataSetFromTximport(txi, colData=design, design=~mouse+ treatment)
dds_treatment_processed(dds_treatment)
treatment_all <- results(dds_treatment_processed, name = "treatment_plus_vs_minus")

Are these equivalent?

deseq2 design • 924 views
ADD COMMENT
2
Entering edit mode
@mikelove
Last seen 1 day ago
United States

Because you are controlling for mouse differences, the genotype effect isn't so easy. What you have above is the genotype effect comparing one mouse vs one mouse. The per-mouse coefficients and the genotype difference is confounded, so you can't extract one without a mixed effects model assuming something about the per-mouse deviations from the group mean. If you need a main effect for genotype, I would re-run a model with ~genotype + treatment, to obtain the genotype effect, and as you say with ~mouse + treatment you can also get the average treatment effect, averaging over all the mice. I like this approach better than averaging the two genotype-specific effects.

ADD COMMENT
0
Entering edit mode

Thanks for that - I think I follow what you are saying about extracting the genotype main effect. I'm less sure about extracting the average treatment effect - wouldn't extracting this from ~genotype + treatment mean you weren't controlling for mouse specific effects (which I think are pretty large)? Isn't the average treatment effect, effectively, like a paired t-test, hence why I wondered about extracting it from ~ mouse + treatment. Is there any particular reason that I shouldn't be using an entirely different design for each question? I can't think what it would be, but it feels wrong.

ADD REPLY
1
Entering edit mode

Yes, you're right, you are better off with ~mouse + treatment for the average treatment effect, I'll edit my post.

ADD REPLY

Login before adding your answer.

Traffic: 891 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6