Question: Need assistance with complicated DESeq2 design
gravatar for alex.pelletier
4.0 years ago by
United States
alex.pelletier10 wrote:

Hello. I'm familiar with DESeq2 but the most complex design equations I've used are having blocked pairs. I need to perform a more complicated design but I'm having trouble formulating the equation. I have a design table that I've read in that contains all the relevant information. Here are the specifications:

condition = sample/control (what I'm testing over)
week = a timepoint comparison of when the sample was generated
grade = a comparison of what grade of disease the sample has (so 4 > 3 > 2 > 1 > 0 ie control)
There are also 3 possible organ sites but some samples are a combination, so I've split it into 3 binary variables: isSiteA, isSiteB, and isSiteC, all of which are 0 or 1

This is what the table looks like:

Sample Condition Week Grade isSiteA isSiteB isSiteC
SAMPA sample 12 3 1 0 0
SAMPB control 2 0 0 1 1
SAMPC control 4 0 0 1 0

Right now my equation is this:
ddsHTSeq <- DESeqDataSetFromMatrix(counts_table, sample_table, design = ~ week : grade : is_GI + is_Liver + is_Skin + condition)

Is this correct at all? I really appreciate any help. FYI I do have more than 3 samples, before anyone asks!

deseq2 design • 506 views
ADD COMMENTlink written 4.0 years ago by alex.pelletier10

hi Alex,

This looks like more of a statistical analysis question than a specific DESeq2 software question (any design or terms you would test in DESeq2 would be the same for a normal linear regression). I'm pressed for time and working on software right now. I'll leave this as a comment and not an answer, so someone else may make suggestions, but if you don't get a response, I'd recommend talking to a local statistician at your institute who could help come up with the design and the terms to test.

ADD REPLYlink written 4.0 years ago by Michael Love26k

I agree. This experimental design is complex enough that you're not going to be able to come up with the "one true design" from a single consultation with a statistician. Rather, you're going to need some back-and-forth dialogue to investigate which factors and interactions are important, and what the best way to model them is. This is why you should definitely find a local statistician who you can keep coming back to for help.

ADD REPLYlink written 4.0 years ago by Ryan C. Thompson7.4k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 324 users visited in the last hour