Question about design formulas in DESeq2 (4 genotypes, 2 batches, 2 cell states)
1
0
Entering edit mode
@rachellcosby-21678
Last seen 4.6 years ago

Hello,

I have a fair amount of experience using DESeq for relatively simple study designs (ie: ~ batch + genotype to look for effect of genotype while controlling for batch) but I am having difficulties deciding on the best formula to address more complicated questions.

Specifically, I am using DESeq to analyze changes in gene expression for 16 samples, with the following design structure:

genotype    batch   immortalized
kko1_new    KKO new yes
kko2_new    KKO new yes
kko1_old    KKO old no
kko2_old    KKO old no
wt1_new WT  new no
wt2_new WT  new no
wt1_old WT  old no
wt2_old WT  old no
P4D7mDox    non new no
P4C11mDox   non new no
P4B3mDox    non new no
P3E12mDox   non new no
P4D7pDox    ind new no
P4C11pDox   ind new no
P4B3pDox    ind new no
P3E12pDox   ind new no

In this case, genotype represents either cells: 1) KO for my gene of interest (KKO) 2) WT for my gene of interest (WT) 3) rescued (non) 4) O.E. (ind)

Ultimately, I want to know how gene expression varies across genotype while controlling for 2 different sources of noise. Ie: How does my GOI regulate gene expression? The first source of noise is batch effects due to experiments (4 samples were prepared initially: "old"; 12 samples were prepared later "new"), and the second is immortalization status. In between "old" and "new" experiments, the KKO cell line underwent a dramatic change in morphology such that I believe there is additional noise that was introduced in addition to variation due to batch effect, which is why I feel that I can't simply resolve this issue by comparing gene expression across "new" samples only.

In this case, would a design formula such as below work, or is it too simple for the question I want to ask? Would it also be beneficial to add an interaction term?

design = ~ batch + immortalized + genotype

Or would it be more appropriate to try an LRT comparison? If so, something like the following?

design = ~batch + immortalized + genotype
gdds_LRT <- DESeq(gdds_LRT, test="LRT", reduced = ~ batch + immortalized)

Any help is greatly appreciated.

deseq2 • 350 views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 9 hours ago
United States

Unfortunately, I don't have much time at this point to provide statistical analysis suggestions, but reserve my time for answering software related questions. We have some description in the vignette on what the meaning of interaction terms are, but for a more in depth conversation regarding whether this makes sense for your experiment, I'd recommend collaborating with a local statistician for guidance on the statistical analysis and design choices.

ADD COMMENT

Login before adding your answer.

Traffic: 703 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6