DESeq2 (Mutant/WT + Two Conditions)

[TheSeqGeek]

1. ) I would like to know whether the level of gene expression is caused by the mutation or by the condition or perhaps there is some mixture. I have three replicates for each experiment.

Condition1 = Exponential Phase
|
|---------> Mutant
|---------------------WT

Condition2 = Stationary Phase
|
|---------> Mutant
|---------------------WT



WT has three replicates in the Exponential phase AND three replicates in the Stationary phase

Mutant has three replicates in the Exponential phase and three replicates in the Stationary phase.


2. ) I already looked over DESeq2 vignette/manual so please don't refer me to that. I would like to know how to setup the factors and levels to make this comparison possible and meaningful to interpret.

Help is greatly appreciated. I know there a few other people who would like to know how to do this.

[dpryan]

Your design is simply:

design <- ~genotype*phase

Depending on how you construct the DESeqDataSet, you would also need to associate your samples to these factors in either the "coldata" (rows are samples and the columns would be "genotype" and "phase") of the 3rd and 4th columns of sampleTable if you're using DESeqDataSetFromHTSeqCount.

[Michael Love]

Yes, Devon has the correct design for you, which is equivalent to

~ genotype + phase + genotype:phase

I would like to know whether the level of gene expression is caused by the mutation or by the condition or perhaps there is some mixture.

In (generalized) linear modeling, this is referred to as an interaction.

If you run DESeq() with the design that Devon provided, you can use the example code below in order to build a results table, where the p-values and adjusted p-values specifies tests of whether the coefficient is zero, and the coefficients available for testing are genotype, phase, and genotype X phase. If there is an effect of the combination which is not explained by simply multiplying the genotype and phase effect, then the interaction term will have a low p-value and adjusted p-value.

you can follow the example in ?results, which is in the 1.4 release version of DESeq2:

# two conditions, two groups, with interaction term
dds <- makeExampleDESeqDataSet(n=100,m=12)
dds$group <- factor(rep(rep(c("X","Y"),each=3),2))

design(dds) <- ~ group + condition + group:condition
dds <- DESeq(dds)

resultsNames(dds)
results(dds, contrast=c("condition","B","A"))
results(dds, contrast=c("group","Y","X"))
# extract the interaction term simply with name
results(dds, name="groupY.conditionB")

[TheSeqGeek]

I got this far

library(DESeq2)

#import the data
directory<-"/Users/Kumka/Documents/R/analysis/"
sampleFiles <- list.files(path="/Users/Name/Documents/R/fnr_analysis/",pattern="*.txt")
sampleFiles


#Give names to experimental conditions
sampleGroup <- factor(rep(rep(c("Mutant","WT"),each=3),2))
sampleCondition <- factor(rep(rep(c("Exponential","Stationary"),each=6),1))
sampleTable<-data.frame(sampleName=sampleFiles, fileName=sampleFiles, group=sampleGroup, condition=sampleCondition)
sampleTable


ddsHTSeq<-DESeqDataSetFromHTSeqCount(sampleTable=sampleTable, directory=directory, design=~group*condition)
dds<-DESeq(ddsHTSeq)
res<-results(dds)

I'm not quite sure how to interpret the log2FC though.... Any thoughts help on this? It's not comparing mutant/wt in one condition but on both is it not?
Here is the code

resultsNames(dds)
group.results <- results(dds, contrast=c("group","WT","Mutant"))
group.results <- group.results[order(group.results$padj),]
head(group.results)
write.csv(group.results, "WT_vs_Mutant_results.csv")

The code

results(ddd)

Is the same as the interaction term same as in the code

results(ddd, name = "groupY.conditonB")

[Michael Love]

When you run a 2x2 analysis with an interaction term, the main effects comparisons, e.g.,

results(dds, contrast=c("group","WT","Mutant"))

generates a results table of log2 WT vs Mutant fold change for the base level of phase.

The log2 WT vs Mutant fold change for the other level of the phase is:

log2 WT vs Mutant + interaction

If the interaction term for a gene is 0, then the WT / Mutant effect is the same for both phases.

[lbragg]

When you run a 2x2 analysis with an interaction term, the main effects comparisons, e.g.,

results(dds, contrast=c("group","WT","Mutant"))

generates a results table of log2 WT vs Mutant fold change for the base level of phase.

The log2 WT vs Mutant fold change for the other level of the phase is:

log2 WT vs Mutant + interaction

If the interaction term for a gene is 0, then the WT / Mutant effect is the same for both phases.

I am using DESeq2 for a similar study design (treatment and time, each with two levels), and need a little help with the above explanation.

Does this code

results(ddd, name = "groupY.conditonB")

give the log2 WT vs Mutant + interaction results?

Also, in this example, how would you get the DE genes between the two levels of phase for the mutant type?

Thanks!

[Michael Love]

hi lbragg,

name = "groupY.conditonB" gives just the interaction effect.

If you want to add the condition B vs A and the interaction effect, you can either provide a numeric contrast (where the numbers correspond to elements of resultsNames(dds)):

results(dds, contrast=c(0,0,1,1))

or if you are using v1.4, you can provide a list to 'contrast':

results(dds, contrast=list(c("condition_B_vs_A","groupY.conditionB"),character()))

See the ?results help for the 'contrast' argument for more details.

Either of these gives the same results table, which is the condition effect for group Y. See here for a primer on how interactions work in linear models: http://en.wikipedia.org/wiki/Interaction_(statistics)