Hi,

I am analysing the dataset with two independent factors - genotype (WT, MUT) and age (pup, adult). I have 4 samples for each condition.

• mouse1 MUT pup
• mouse2 MUT adult
• mouse3 WT pup
• mouse4 WT adult etc.

What I want to check is how being the MUTANT affects gene expression across the age and if the impact of mutation is age dependent.

What I need is to build a model that shows me:

1. how age affects the counts
2. how genotype affects the counts
3. and how the interaction age*genotype affects the counts - is it significant?

However after reading on forums I am confused and I am not sure which approach is best. Also I am not sure if I should use LRT and in which step.

My first idea is the model with interaction, as a 'baseline' I set Wild Type adult:

dds <- DESeqDataSetFromMatrix(countData=countData, 
                              colData=metaData, 
                              design= ~genotype+age+genotype:age, 
                              tidy = TRUE)
dds
dds$genotype
dds$genotype = relevel( dds$genotype, "wt")
dds$genotype

dds$age
dds$age = relevel( dds$age, "adult")
dds$age

keep <- rowSums(counts(dds)) >= 10
dds <- dds[keep,]

dds <- estimateSizeFactors(dds)
dds <- DESeq(dds)

resultsNames(dds)

Or is it better to use 'group' design? What is more informative in my case?

dds$group <- factor(paste0(dds$genotype, dds$age))
design(dds) <- ~ group
dds <- DESeq(dds)
resultsNames(dds)

resultsNames(dds) [1] "Intercept" "group_mutpup_vs_mutadult" "group_wtadult_vs_mutadult" [4] "group_wtpup_vs_mutadult"

results(dds, contrast=c("group", "wtpup", "mutpup"))
results(dds, contrast=c("group", "wtadult", "mutadult"))
results(dds, contrast=c("group", "wtpup", "mutadult"))

In the group design how do I interpret the last contrast - is it the same as interaction from the previous model? I will be grateful for any advice, this is the first time I am trying multifactorial analysis in DESeq2.

Hey, these are somewhat vague in their definition: "how age affects the counts", "how genotype affects the counts". To attempt to answer what [I believe] you want here, I would consider to just run an independent model of the form ~ age + genotype (or even just ~ age and ~ genotype separately) and then check the coefficient for each.

From your interaction design, ~ genotype + age + genotype:age, the interaction term, age:genotype, can be checked via a coefficient named "agepup.genotypemut".

There is useful information accessible via ?DESeq2::results() (at the bottom) about 2 x 2 interaction designs. I will paste here:

## Example 2: two conditions, two genotypes, with an interaction term

     dds <- makeExampleDESeqDataSet(n=100,m=12)
     dds$genotype <- factor(rep(rep(c("I","II"),each=3),2))

     design(dds) <- ~ genotype + condition + genotype:condition
     dds <- DESeq(dds) 
     resultsNames(dds)

     # the condition effect for genotype I (the main effect)
     results(dds, contrast=c("condition","B","A"))

     # the condition effect for genotype II
     # this is, by definition, the main effect *plus* the interaction term
     # (the extra condition effect in genotype II compared to genotype I).
     results(dds, list( c("condition_B_vs_A","genotypeII.conditionB") ))

     # the interaction term, answering: is the condition effect *different* across genotypes?
     results(dds, name="genotypeII.conditionB")

So, you can gauge age and genotype affects via the interaction design, but always with respect to each other, for example:

• age effect for mut genotype
• age effect for wt genotype
• genotype effect for pup
• genotype effect for adult
• et cetera

You can possibly check for general age and genotype effects via ANOVA / LRT with the interaction design, but one needs to be very careful about the interpretation. It may be better to consult with a local statistician.

.

Regarding the group design, I would not say that it is the same as the interaction design, or that any coefficient among the group design coefficients are the same as those in the interaction design. The group design, literally, just permits pairwise comparisons between any 2 groups.

Kevin

results(dds, contrast=c("group", "wtpup", "mutadult"))

That contrast doesn't make a lot of sense, you wouldn't do that.

Imagine taking the results from here:

results(dds, contrast=c("group", "wtpup", "mutpup"))
results(dds, contrast=c("group", "wtadult", "mutadult"))

And for each gene, subtracting the fold change of one from the other, to find the genes where the fold change caused by the mutation is different in pups as opposed to adult. You'd then get a difference of fold changes, but no p-value.

If you do the design with interactions, and this contrast

results(dds, name="genotypeII.conditionB")

you will get what I described with p-values. That's why you would do interactions.