DESeq2 multi-factor designs

id0 · 01-09-2014, 03:35 PM

I am getting started with DESeq2. I am trying to understand how multi-factor design works. This is fairly basic question, but I can't find the proper answer.

From the vignette:

Code:

We can account for the different types of sequencing, and get a clearer 
picture of the differences attributable to the treatment. As condition is the 
variable of interest, we put it at the end of the formula. Here we

design(dds) <- formula(~ type + condition)
dds <- DESeq(dds)

So how exactly are the two parameters treated? What do the significant genes in the results mean? Are they the ones that are changing in condition regardless of type, or the ones that are changing in both type and condition? What happens when you have three or more parameters?

Additionally, although results for condition are output by default, results for type can be retrieved as well. Does the order or parameters only matter for the output or is the test performed differently?

dpryan · 01-10-2014, 01:46 AM

The normal phrasing for that would be "differences due to type when controlling for condition" for the "type" results and "differences due to condition when controlling for type" for the "condition" results. The syntax is the same as is used for other linear and generalized linear models in R, so that's why it's not described in depth. If "condition" has levels "A", "B", and "C", then (generally) "A" will be treated as the baseline level and you get statistics for the "BvsA" and "CvsA" comparisons (I think the most recent version of DESeq2 implements contrasts, so you can then test whatever you really want).

id0 · 01-10-2014, 07:22 AM

Quote:

Originally Posted by dpryan

The normal phrasing for that would be "differences due to type when controlling for condition" for the "type" results and "differences due to condition when controlling for type" for the "condition" results. The syntax is the same as is used for other linear and generalized linear models in R, so that's why it's not described in depth. If "condition" has levels "A", "B", and "C", then (generally) "A" will be treated as the baseline level and you get statistics for the "BvsA" and "CvsA" comparisons (I think the most recent version of DESeq2 implements contrasts, so you can then test whatever you really want).

That was very helpful.

I think you misunderstood the part of my question regarding multiple parameters, though. What happens when the design goes from formula(~ x + y) to formula(~ x + y + z)?

dpryan · 01-10-2014, 07:25 AM

Ah, with more parameters things are the same. You're always testing one thing while accounting for the others.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
DESeq2: diff gene expression between species using gene-specific normalization factor	mra	Bioinformatics	4	12-01-2014 07:17 AM
DESeq Multi-factor designs: determining the significance of model terms	DJParker	Bioinformatics	8	07-21-2014 03:21 PM
DESeq2: Multi-factor designs	sindrle	Bioinformatics	10	10-21-2013 08:47 AM
Splice variants from a multi factor RNA-Seq	cfreije	Bioinformatics	3	08-26-2013 02:27 AM
DEXSeq for multi-factor design	alittleboy	Bioinformatics	5	06-27-2013 10:43 AM

01-09-2014, 03:35 PM	#1
id0 Senior Member Location: USA Join Date: Sep 2012 Posts: 130	DESeq2 multi-factor designs I am getting started with DESeq2. I am trying to understand how multi-factor design works. This is fairly basic question, but I can't find the proper answer. From the vignette: Code: We can account for the different types of sequencing, and get a clearer picture of the differences attributable to the treatment. As condition is the variable of interest, we put it at the end of the formula. Here we design(dds) <- formula(~ type + condition) dds <- DESeq(dds) So how exactly are the two parameters treated? What do the significant genes in the results mean? Are they the ones that are changing in condition regardless of type, or the ones that are changing in both type and condition? What happens when you have three or more parameters? Additionally, although results for condition are output by default, results for type can be retrieved as well. Does the order or parameters only matter for the output or is the test performed differently?

01-10-2014, 01:46 AM	#2
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,480	The normal phrasing for that would be "differences due to type when controlling for condition" for the "type" results and "differences due to condition when controlling for type" for the "condition" results. The syntax is the same as is used for other linear and generalized linear models in R, so that's why it's not described in depth. If "condition" has levels "A", "B", and "C", then (generally) "A" will be treated as the baseline level and you get statistics for the "BvsA" and "CvsA" comparisons (I think the most recent version of DESeq2 implements contrasts, so you can then test whatever you really want).

01-10-2014, 07:25 AM	#4
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,480	Ah, with more parameters things are the same. You're always testing one thing while accounting for the others.