DESeq2 design question - Two group treatment differences accounting for response
1
0
Entering edit mode
jhutchin • 0
@jhutchin-11223
Last seen 7.0 years ago

I've got a DESeq2 design matrix question.

I have two groups of mice, young and old.

Mice in both groups have been given of 3 different treatments  (control treatment (treatment zero), and two treatments of interest (treatment1 and treatment2). Some of the treated mice show a response to the treatment, some do not. The control treatment mice never show a response. Some of the old mice do not respond to one treatment.

I'd like to pull out:

1) the DEGs after treatment1 compared to treatment0 in young mice specific to the repsonse
2) the DEGs after treatment2 compared to treatment0 in young mice specific to the repsonse
3) the DEGs after treatment1 compared to treatment0 in old mice specific to the repsonse
4) the DEGs after treatment1 compared to treatment0 in old mice specific to the response
1) the differences between 1) and 3) (i.e. some sort of interaction term)
1) the differences between 2) and 4) (i.e. some sort of interaction term)

I know how to do all of this without addressing the response (i.e. as a simple interaction term based design of ~age+treatment +age:treatment) but I have no clue how to address the response variable and  get only genes that are changing in the the responding mice.

Is there a way to do this all in one design or should I be looking to split it up somehow?

 

Here's the metadata:

age treatment response
Old treatment0 no_response
Old treatment0 no_response
Old treatment0 no_response
Old treatment0 no_response
Old treatment0 no_response
Old treatment0 no_response
Old treatment0 no_response
Old treatment1 no_response
Old treatment1 no_response
Old treatment1 no_response
Old treatment1 no_response
Old treatment1 no_response
Old treatment1 no_response
Old treatment1 no_response
Old treatment1 no_response
Old treatment2 no_response
Old treatment2 no_response
Old treatment2 no_response
Old treatment2 no_response
Old treatment2 no_response
Old treatment2 no_response
Old treatment2 reponse
Old treatment2 reponse
Young treatment0 no_response
Young treatment0 no_response
Young treatment0 no_response
Young treatment0 no_response
Young treatment0 no_response
Young treatment0 no_response
Young treatment0 no_response
Young treatment1 no_response
Young treatment1 no_response
Young treatment1 no_response
Young treatment1 no_response
Young treatment1 no_response
Young treatment1 reponse
Young treatment1 reponse
Young treatment1 reponse
Young treatment1 reponse
Young treatment1 reponse
Young treatment1 reponse
Young treatment1 response
Young treatment2 no_response
Young treatment2 no_response
Young treatment2 no_response
Young treatment2 no_response
Young treatment2 reponse
Young treatment2 reponse
Young treatment2 reponse
Young treatment2 reponse
Young treatment2 reponse
Young treatment2 reponse

 

Thanks,
John

 

 

 

deseq2 interaction model multiple factor design multifactorial design • 1.7k views
ADD COMMENT
0
Entering edit mode

No, the responder status is a separate phenotype (whether the tumor regresses (response) after treatment or not (no_response))

ADD REPLY
0
Entering edit mode

Hmm, there is a bit of an issue here in splitting the samples based on a random effect. I see what you want to do though. I mean a practical approach, not dealing with the random effect problem, would be to have treatment with levels: control, trt1no, trt1yes, trt2no, trt2yes. There will be some fixing to do here, because you don't have old mice who are responders to trt1, so you can't fix that interaction term with age. You will have to remove such terms from the model.matrix manually (see vignette for code on that).

You could also think about using mixed effects models, but this isn't built in to DESeq2 obviously.

ADD REPLY
0
Entering edit mode

So, in general you're suggesting making a custom matrix, dropping out columns that are unrepresented in the metadata and then making contrasts to get at what I'm interested in?

ADD REPLY
0
Entering edit mode

After recoding the `treatment` variable, yes. From scanning, I think it's just one interaction that will have no samples to fit it, but you'll want to take a look at the model matrix.

ADD REPLY
0
Entering edit mode

Great! I should be good to go from here. Thanks!

ADD REPLY
0
Entering edit mode
@mikelove
Last seen 8 hours ago
United States

And to clarify, do you determine the responder status using the RNA-seq data?

ADD COMMENT
0
Entering edit mode

 

No, the responder status is a separate phenotype (whether the tumor regresses (response) after treatment or not (no_response))

ADD REPLY

Login before adding your answer.

Traffic: 489 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