Can I control the exposure time as a batch effect?

Question

[greyman]

Hi, the experiment that I conducted consists of a control and 4 different concentrations of chemical treatments (5 replicates per treatments) under two different exposure time (12 hours and 48 hours). As I was interested to compare the expression profile between the two time points, I found out there is a batch effect caused by sequencing both time points at different batches from my PCA plots: the 12hours group has separated entirely from the 48 hours group, even for my two control groups they were also far far away from each other.

Can I therefore treat the exposure time as a batch effect in my DESeq design model? What disadvantage would there be the batch is exactly same as the exposure time? I couldn't find similar posts or suggestion as I had searched in google. Here I would like to treat the chemical concentration as a linear model

dds <- DESeqDataSetFromMatrix(counts, metadata, design = ~time + chemical.con )

DESeq(dds)

resultsNames(dds)
[1] "Intercept" "time"      "chemical.con"

Something doesn't look right here, seems like I couldn't compare the expression profiles between 8 days and 4 days, any advice?

Thank you very much

[rpolicastro]

Can you provide a picture of the PCA plot? It's not clear from just your description alone whether you have a batch effect or not.

Answer 1

[Devon Ryan]

Can I therefore treat the exposure time as a batch effect in my DESeq design model?

Yes, treating time as a batch effect seems reasonable if your control samples are showing notable effects due to it.

What disadvantage would there be the batch is exactly same as the exposure time?

Well, you can't measure the effect of time, since it's convolved in the batch effect.