Question: DESeq2 question about design formula with nested variables
0
8 days ago by
Bash0
Bash0 wrote:

Hi,

I am currently working on an experiment and have some trouble with the design. I've looked on the bioconductor help pages and the manual, but I have the feeling that I am overlooking something simple. Maybe you could help with some insights, or point me to a post with a similar setup.

My setup is as follows:

genotype   timepoint  flexible
GT1        T1         flex
GT3        T1         flex
GT2        T1         non_flex
GT4        T1         non_flex
GT5        T1         non_flex
GT1        T2         flex
GT3        T2         flex
GT2        T2         non_flex
GT4        T2         non_flex
GT5        T2         non_flex
GT1        T3         flex
GT3        T3         flex
GT2        T3         non_flex
GT4        T3         non_flex
GT5        T3         non_flex


I have 5 different genotypes (GT1..GT5) which are genotypically different. Each of these genotypes have been put in the cooler at T2 and after some time RNA was extracted. Later, the samples were put out cooler and after some time RNA was extracted (T3). Every genotype is either flexible or non-flexible (i.e. flexible means in this case they deal better with cold). For each of the sample/timepoint I have three biological replicates (omitted from the scheme above).

With this experiment I want to know which genes change over the different time points and what the difference is between genotypes while taking in account the flexible trait.

What I could do is making a design like this (which follows the example of the time series experiment found in the manual):

~ genotype + timepoint + genotype:timepoint

In which the genotype is modelled with the interaction over time. However in this case I don't take in account the flexible trait. With a result-set like that, I could easily calculate the overlap between between different interaction terms manually (i.e. (GT1T2/GT1T1) vs (GT2T2 /GT2T1)), but this doesn't seem the best way to go.

What I also could do is is replacing genotype for flexible in the formula, which would make the design simpler but it will not take in account the differences between genotype, which is something I don't want. Do you have any tips to solve this?

Best regards,

Bas

deseq2 • 54 views
modified 8 days ago by Michael Love24k • written 8 days ago by Bash0
0
8 days ago by
Michael Love24k
United States
Michael Love24k wrote:

If you think the genotype differences are important at baseline but the interaction effect is only different between flex and unflex then you should be able to adapt the technique outlined in the vignette about controlling for individual baselines while testing for group differences.