Question

DESeq2 Design Formula for Time Series Experiment

1

Entering edit mode

gkuffel ▴ 10

@gkuffel-12871

Last seen 5.4 years ago

Loyola University

Hi everyone,

I have miRNA-Seq data for an experiment with the following parameters:

SampleID Treatment Timepoint Patient

P1_0 TB 0 1

P1_1 TB 1 1

P1_2 TB 3 1

HC1_0 None 0 Control1

HC1_2 None 3 Control1

P2_0 TB 0 2

P2_1 TB 1 2

P2_2 TB 2 2

P2_3 TB 3 2

HC2_0 None 0 Control2

HC2_3 None 3 Control2

I am really struggling figuring out how to enter the design formula. There are 2 patients each with a baseline sample before treatment and 2 samples at time points after treatment. Each patient has an age-matched control with samples taken at two time points. One patient is a full responder to treatment and one is not but I am blinded to that info. The investigator wants to identify DEmiRNAs in the following:

1. Between patients

2. As a result of treatment over time

Do I need to include interaction terms? So far I have identified 128 DEmiRNAs using this formula:

design=~timepoint + individual

Which I believe is testing for differences between individuals across all time points, can anyone provide some much needed guidance?

deseq2 • 996 views

ADD COMMENT • link updated 6.7 years ago by David Requena ▴ 20 • written 6.7 years ago by gkuffel ▴ 10

score 2 · Accepted Answer · 2017-09-22

2

Entering edit mode

David Requena ▴ 20

@david-requena-13990

Last seen 6.3 years ago

The Rockefeller University. New York, U…

To compare time points, you should consider "timepoint" in your formula, as you did.

To compare treated vs untreated, you should include "treatment" in your formula.

If you think that your treated patients will experience changes over time, which are different than the changes experienced by the untreated, you should add the interaction term "treatment:timepoint".

I guess that "individual" is your column "patient" or "treatment:age", which are equivalent.

I think you'll might find the explanation in the DEseq2 vignette useful (section "Interactions"):

https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#interactions

And also these posts:

A: DESeq2 paired multifactor test

With DESeq2 "Not full rank" Error with design ~ line + time + condition

David

ADD COMMENT • link 6.7 years ago David Requena ▴ 20

0

Entering edit mode

David,

Thank you so much for your reply. I have a few more questions if you have time:

1. So with my current design (design =~ Timepoint + Patient) am I testing for differences between patients while controlling for variability caused by time?

2. If I use the design: design = ~ Timepoint + Treatment, is that testing for differences between treatments while controlling for variability caused by time?

3. When you ask if I think the treated patients will experience changes over time different than changes experienced by untreated do you mean changes caused by the treatment or just naturally ocurring changes?

4. Last and maybe most important question. The most important comparsison the group wants to make is does the treatment have an effect on the patients over time. I still don't understand if I am using the correct design for this.

ADD REPLY • link 6.6 years ago gkuffel ▴ 10