DESeq2 Proper Design

Question

[KyleBos]

Hello Biostars

I have a quick question about the design parameter for DESeq2 analysis

I have an experiment where I have 6 individual animals and the following timepoints: Day 0 (baseline and infection date), Day 1, Day 3, Day 7, Day 10, Day 14. Each animal had a sample drawn on every timepoint and RNAseq run for each timepoint in triplicates.

I am measuring gene changes from Day 0 to each of the corresponding timepoints.

For design I have been just using

design = ~ timepoint

since there are no treatments, just the infection date and subsequent measurements. Would this be correct? The animals were all housed together and were matched by age, weight, and I believe similar genotype.

Thank you!

Kyle

[h.mon]

and RNAseq run for each timepoint in triplicates.

You mean for each time point you prepared three libraries? Or you split the same libraries over three lanes? Both are technical replicates, but I don't see the point of preparing three libraries.

You really don't have a control group? A group of animals of matched age, weight, genotype, but not infected? How will you be able to say differences are due to infection and not just aging? Or maybe there was a thunderstorm one day and the animals got really stressed that day. Or someone was listening to loud death metal and stressed the animals. You need a control group, subjected to the same conditions except for the one you are interested, in this case, infection challenge.

[KyleBos]

I believe it was one library over three lanes.

As for the control group, unfortunately this experiment was designed and run well before I joined this lab so there was no control group. This is my lab's first foray into the RNAseq world, we mainly work with flow cytometry and functional assays that can be analyzed using flow. I do not know who or if they consulted with anybody to design this experiment but I agree that there should have been a control group.

edit: I don't know if it's relevant but the experiment was run using sorted Natural Killer cells since we are trying to find transcriptomic signature of NK cells during the course of SIV infection.

[Kevin Blighe]

If you're just interested in differences between Day 0 and each other timepoints, then using ~ timepoint is fine. If you additionally expect difference across animal, you may want to include ~ timepoint + animal.

After you normalise and transform the data, you should check the dispersion plot, the symmetrical heatmap, and the PCA bi-plot to see if any problems may exist, such as outliers.

If you're still in doubt, then be aware that DESeq2 developer, Michael Love, is more active on the Bioconductor support site: https://support.bioconductor.org/