Time series RNA seq design query

Question

[kra277]

Hello,

I am working with Deseq2 on RNA seq time-series data. I have RNA seq data at baseline before any intervention (timepoint0), RNA seq data with two different interventions (Placebo, and Condition) at TP1 and TP2. I went through the time-series tutorial for Deseq2 and don't think the same example applies to my dataset. Below is the sample info

# Phenotypes (minimal reprex)
colData <- data.frame(
        "patient" = c(rep(c("sam_1", "sam_2", "sam_3", "sam_4", "sam_5", "sam_6"), 3)), 
        "timepoint" = c(rep("tp0", 6), rep("tp1", 6), rep("tp2", 6)), 
        "treatment" = c(rep("none", 6), rep(c("placebo","condition"), 6))
     )

# Here is the deseq2 formula I was thinking to try but not sure if its correct
ddsMat <- DESeqDataSetFromMatrix(countData = countData,
                             colData = colData, 
                             design = ~ treatment + timepoint + treatment:timepoint)

# Likelihood ratio test to get genes that have shown to be DE over time
deseq_res <- DESeq(ddsMat, test="LRT", reduced = ~  treatment + timepoint)

I don't have any treatment at time point 0 so I am not sure if I can use the above formula. Could you please point me in the right direction, please?

[swbarnes2]

What exact comparison are you trying to do? I'm not sure that will even work, since all the t0's have the same Treatment. Are you sure you want to do an LRT, and not just simple comparisons?

[kra277]

Hi, Thank you for your response,

I want to check differentially expressed genes over time between the treatments. I did try simple comparisons for each treatment, tp0 vs tp1 and tp0 vs tp2. I was trying to see if there is a design that shows genes that are different over time in between the treatments (placebo and drug).

[swbarnes2]

It's just that the results from LRT can be hard to interpret. You've only got three time points, you can just do the three pairwise interactions.

[kra277]

Okay. Thank you. I will do the individual analysis

[ATpoint]

You can also do a full-factorial design, so add an additional column like group=paste(timepoint, treatment, sep="_") and then do a design ~ group. That way you can easily contrast everything to everything as you like, and then visualize the combined DEGs in a heatmap to find patterns. The advantage of this is that you can do the Wald test which actually tests fold changes rather than "only" the goodness of fit as in the LRT which at least for me (no formal stats background) has always been hard to interpret.

[kra277]

Thank you for your time and response. I will try this approach.