Dear All,

Hope you are well. I am running time course experiment to examine differentially expressed genes from tumour tissues between two conditions, radiotherapy and sham-radiotherpapy groups at 4 time points post radiotherapy. Each time point had size matched untreated controls.

I am examining both the differentially expressed genes at specific time (all 4 time points). I also investigated differentially expressed genes overtime by adding time as another feature in the model and effect modification (interaction) between time and Treatment condition (treated/untreated).

I understand DESEq2 is based on negative binomial (Gamma poison). I am assuming how the model work in the context of gene expression is to pre-calcuate the log-fold changes of genes between two groups and that will be used to fit negative bionomial and then perfrom a test whether the logfold change is significant. Am I right?

In a setting of time-course, how does the log-fold changes are calculated in the presence of Time and Time:Treatment interaction? I am getting confused with it.

I generally got around 1000 genes that were differently expressed at specific time (just pairwise comparsion). However, once I fitted the model by adding Time and interaction between Time and Treatment condition in the model. I got around 3000 genes that were differentially expressed. Does this sound right to you? I should expect less genes that are differentially expressed overtime than just two group comparision at specific time point? What do you think? Following is my model design matrix for time course. Looking forward to hearing from you soon.

dds <- DESeqDataSetFromMatrix(countData = Raw.Count, 
                              colData = metadata,
                              design = ~ Treatment + Time + Treatment:Time)
# DESeq
dds<- DESeq(dds)
# Diff
RT_vs_sham.RT<- results(dds, contrast=c("Treatment","Radiotherapy", "Sham-radiotherapy"), alpha = 0.05)
summary(RT_vs_sham.RT)

Kind Regards,

Synat,

For questions about how to design the statistical analysis and interpret the results, I recommend collaborating or consulting with a statistician or someone familiar with linear models in R. I have to reserve my time on the support site for software related questions.