Dear Michael,

I would like to start by thanking you for the DESeq2 package and the support you give. Its remarkable.

I have a doubt in the design formula that I would kindly ask your opinion on.

In my experiment, I have 2 variables: 1) Condition = infected cells and non-infected cells 2) time = I took samples 0, 24 and 48 hours after infection

I am interested in the effect of the infection on the culture along time. As I am not 100% sure if the initial infected sample or the associated non-infected sample per timepoint are better comparisons, I saw in your vignette that one can do a full time-series setup. Therefore I ran the following design:

DESeq_data <- DESeqDataSetFromHTSeqCount(
  sampleTable = sample_information,
  directory = count_dir,
  design = ~ condition + time + condition:time)

colData(DESeq_data)$condition <- factor(colData(DESeq_data)$condition,
                                        levels = c("inf", "non"))

# Set non-infected as comparing condition
DESeq_data$condition <- relevel(DESeq_data$condition, "non")

DESeq_data <- DESeq(DESeq_data)

# Perform likelihood ratio test to remove condition-specific differences over time
DESeq_data <- DESeq(DESeq_data, test="LRT", reduced = ~condition + time)

res48inf <- results(DESeq_data, independentFiltering = T, name="conditioninf.time48", test= "Wald", lfcThreshold = 0)
res24inf <- results(DESeq_data, independentFiltering = T, name="conditioninf.time24", test= "Wald", lfcThreshold = 0)

My questions are now the following: 1) Does this setup make sense to evaluate the effect of the infection along time, or would it be better to make single comparisons (infected vs non-infected / infected 24vs0 etc)? 2) Giving this design, I can only make comparisons for the effect of infection along time for 24 vs 0 and 48 vs 0. Is there a possibility to compare 48 vs 24 in this setup, or would I need to do that separate?

Many thanks in advance for your help and consideration.

Kind Regards, Nikolaus

You can use the full design to make comparisons at each time point -- see the time series example in the rnaseqGene workflow.

You can compare 48 to 24 from contrasts of coefficients, using the contrast argument -- I recommend to find a collaborator familiar with linear models in R to help you form these contrasts and interpret results.