Is my time course analysis with DESeq2 valid?
1
0
Entering edit mode
Jason • 0
@758cc442
Last seen 4 days ago
Austria

Hey community!

As a pure behavioural ecologist who has stumbled into the world of gene expression analysis and am a novice in analyzing it, I am asking for help in validating whether my model is correct for the type of analysis I want to do. I have been reading the DESeq2 manual and trying to figure out the correct way to analyze my data.

My experiment is about finding differential gene expression in ants facing a pathogen infection, so I have samples from three timepoints (Time: 0, 1, 2) and a condition (Treatment: control, infected). I am using colony fragments from a shared original nests for both the conditions over all timepoints, meaning I should control for a batch effect for the model (original nest, in the model as Replicate).

I approached the analysis as follows:

dds <- DESeqDataSetFromMatrix(countData = tempdf,
                              colData = sfile,
                              design=~ Replicate + Treatment + Time + Treatment:Time)
dds <- DESeq(dds)

Using contrasts, I wanted to specifically focus on differential expression over timepoint 1 and timepoint 2, which I got from the interaction term Treatment:Time

resT1 <- results(dds, name="TreatmentInfected.Time1", test = "Wald")
resT2 <- results(dds, name="TreatmentInfected.Time2", test="Wald")

It was a bit unclear to me whether I should use these types of comparisons or whether to include a reduced model with an LRT test, or whether the interaction model would be the correct way to go.

The results of these models seem logical enough, quite in line with a quick and dirty look at the expression counts, and finding with quite good accuracy the same differentially expressed genes as an attempt with EdgeR (however I could not get the batch effect to work there for some reason).

I would be extremely greatful for any comments on whether my approach is correct or whether I should modify my model and take another approach!

GeneExpression DESeq2 TimeCourse • 189 views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 6 days ago
United States

For choosing the correct statistical design, I recommend working with a local statistician or someone familiar with linear models in R. I have to reserve my time on the support site for software-related questions.

You may also try out ExploreModelMatrix:

https://csoneson.github.io/ExploreModelMatrix/

ADD COMMENT

Login before adding your answer.

Traffic: 664 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