SEQanswers

Go Back   SEQanswers > General



Similar Threads
Thread Thread Starter Forum Replies Last Post
starBase: explore microRNA–mRNA interaction maps from CLIP-Seq and Degradome-Seq data yjhua2110 Bioinformatics 8 09-27-2011 08:45 AM
starBase: explore microRNA–mRNA interaction maps from CLIP-Seq and Degradome-Seq data yjhua2110 Literature Watch 1 02-02-2011 08:50 AM
ChIP-Seq: Integrative analysis of genomic, functional and protein interaction data pr Newsbot! Literature Watch 0 11-27-2010 03:02 AM
ChIP-Seq: Quantitative Interaction Proteomics and Genome-wide Profiling of Epigenetic Newsbot! Literature Watch 0 09-21-2010 02:00 AM
Can DESeq and edgeR deal with in-balanced RNA-seq data? asiangg Bioinformatics 3 05-26-2010 04:45 AM

Reply
 
Thread Tools
Old 07-13-2011, 11:50 AM   #1
EMeyer
Junior Member
 
Location: Austin TX

Join Date: Jul 2008
Posts: 7
Default DESeq: GLMs for RNA-Seq with interaction terms

I have a reasonably large RNA-Seq dataset for a non-model plant, and want to fit a GLM with interaction terms:
expression ~ treatment + time + treatment:time

In theory, I should be able to test the significance of each term by comparing a reduced model (dropping that term) with the full model above.

It seems DESeq can handle this for main effect terms: it produces a vector of p-values with a reasonable distribution. See this example, using data from the pasilla package:
data(pasillaGenes)
design <- pData(pasillaGenes)[, c("condition", "type")]
fullCountsTable <- counts(pasillaGenes)
cdsFull <- newCountDataSet(fullCountsTable[1:1000,], design)
cdsFull <- estimateSizeFactors(cdsFull)
cdsFull <- estimateDispersions(cdsFull, method="pooled")
fit1 <- fitNbinomGLMs(cdsFull, count ~ type + condition)
fit0 <- fitNbinomGLMs(cdsFull, count ~ type)
pvalsGLM <- nbinomGLMTest(fit1, fit0)


However, when I try this with an interaction term included it gives clearly wrong results: every p value is either 0 or 1.
fit1 <- fitNbinomGLMs(cdsFull, count ~ type + condition + type:condition)
fit0 <- fitNbinomGLMs(cdsFull, count ~ type + type:condition)
pvalsGLM <- nbinomGLMTest(fit1, fit0)

So my questions:
(a) Is this even possible in DESeq?
(b) If so, can anyone spot the error in how I coded this?

Thanks for any advice!
EMeyer is offline   Reply With Quote
Old 07-13-2011, 10:23 PM   #2
Simon Anders
Senior Member
 
Location: Heidelberg, Germany

Join Date: Feb 2010
Posts: 994
Default

Hi

you test for a rather odd contrast:
Quote:
Code:
fit1 <- fitNbinomGLMs(cdsFull, count ~ type + condition + type:condition)
fit0 <- fitNbinomGLMs(cdsFull, count ~ type + type:condition)
The standard ANOVA or ANODEV approach is to test first for main effects without any interaction terms present in the fit, and then for the interaction, with all main effect present in both fits, i.e.,

Code:
fit1 <- fitNbinomGLMs(cdsFull, count ~ type + condition + type:condition)
fit0 <- fitNbinomGLMs(cdsFull, count ~ type + condition)
For the pasilla data, this shouldn't give many hits, though. (If there were interactions, this would be rather worrying.)
Simon Anders is offline   Reply With Quote
Old 07-21-2014, 02:10 PM   #3
raphael123
Member
 
Location: Mc Gill -- Montreal

Join Date: Dec 2013
Posts: 37
Default

Hi Simon !
Usually each coeficient in a GLM is associated with a p value, do you know how to get these p value from the GLM of DEseq ?
raphael123 is offline   Reply With Quote
Reply

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off




All times are GMT -8. The time now is 04:32 AM.


Powered by vBulletin® Version 3.8.9
Copyright ©2000 - 2020, vBulletin Solutions, Inc.
Single Sign On provided by vBSSO