correcting for gene-length and GC-content bias in DESeq2

Golsheed · 01-19-2015, 10:55 AM

Hi,
I was wondering whether DESeq2 has a feature to correct for gene-length and GC-content bias simultaneously?
I currently am correcting for GC-content bias using the method explained in DESeq2 manual (i.e., the normalizationFactors matrix). But I need to do correction for gene-length, since I'm trying to find differentially expressed genes between three species. Assuming that L_ij is the length of gene i in condition (subgroup) j, how do I incorporate matrix L along with the normalizationFactors matrix into the data set?
I'd appreciate any help.
Thanks,
Golsheed

Michael Love · 01-19-2015, 12:43 PM

Are you using CQN? Something to think about is that CQN estimates the GC bias assuming that the genes are the same length across samples. See ?cqn.

What's the distribution of gene length differences, e.g.:

summary(log2(L[,1] / L[,2]))

Golsheed · 01-19-2015, 12:52 PM

Thanks for the answer. EDASeq is used for that part of the analysis to correct for GC bias.

Golsheed

Golsheed · 01-19-2015, 12:56 PM

Can I correct for both GC content and gene length with EDASeq and then use the output from EDASeq in DESeq2?

Golsheed

Michael Love · 01-19-2015, 03:09 PM

That's a question for the EDASeq devels. DESeq2 can take any normalization matrix.

Note that, similar to CQN, the covariates (gene length, GC) are provided as vectors to EDASeq, i.e. one per gene, not a matrix of gene lengths.

Golsheed · 01-19-2015, 03:25 PM

Thanks a lot. Your response is extremely helpful.

Golsheed

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Gene length and GC content	thaarapillai@gmail.com	Bioinformatics	0	06-10-2013 04:15 PM
RNA-seq and GC content bias	federete	RNA Sequencing	3	03-01-2013 06:24 AM
explain unimodal GC-content bias	shuteo	Illumina/Solexa	2	08-29-2012 12:43 PM
PubMed: Correcting for cancer genome size and tumour cell content enables better esti	Newsbot!	Literature Watch	0	08-28-2012 02:00 AM
RNA-Seq: Summarizing and correcting the GC content bias in high-throughput sequencing	Newsbot!	Literature Watch	0	02-11-2012 02:00 AM

01-19-2015, 10:55 AM	#1
Golsheed Member Location: Montreal Join Date: Oct 2014 Posts: 49	correcting for gene-length and GC-content bias in DESeq2 Hi, I was wondering whether DESeq2 has a feature to correct for gene-length and GC-content bias simultaneously? I currently am correcting for GC-content bias using the method explained in DESeq2 manual (i.e., the normalizationFactors matrix). But I need to do correction for gene-length, since I'm trying to find differentially expressed genes between three species. Assuming that L_ij is the length of gene i in condition (subgroup) j, how do I incorporate matrix L along with the normalizationFactors matrix into the data set? I'd appreciate any help. Thanks, Golsheed

01-19-2015, 12:43 PM	#2
Michael Love Senior Member Location: Boston Join Date: Jul 2013 Posts: 333	Are you using CQN? Something to think about is that CQN estimates the GC bias assuming that the genes are the same length across samples. See ?cqn. What's the distribution of gene length differences, e.g.: summary(log2(L[,1] / L[,2]))

01-19-2015, 12:52 PM	#3
Golsheed Member Location: Montreal Join Date: Oct 2014 Posts: 49	Thanks for the answer. EDASeq is used for that part of the analysis to correct for GC bias. Golsheed

01-19-2015, 12:56 PM	#4
Golsheed Member Location: Montreal Join Date: Oct 2014 Posts: 49	Can I correct for both GC content and gene length with EDASeq and then use the output from EDASeq in DESeq2? Golsheed

01-19-2015, 03:09 PM	#5
Michael Love Senior Member Location: Boston Join Date: Jul 2013 Posts: 333	That's a question for the EDASeq devels. DESeq2 can take any normalization matrix. Note that, similar to CQN, the covariates (gene length, GC) are provided as vectors to EDASeq, i.e. one per gene, not a matrix of gene lengths.

01-19-2015, 03:25 PM	#6
Golsheed Member Location: Montreal Join Date: Oct 2014 Posts: 49	Thanks a lot. Your response is extremely helpful. Golsheed