Cufflinks, differentially expressed genes

statsteam · 02-01-2012, 12:09 PM

Hi,

I am trying to run edgeR or DEGseq using the output from cufflinks.
I usually use mapped reads count as an input to edgeR or DEGseq. What cufflinks output do I need to use for an input to edgeR or DEGseq? I am thinking about adding "coverage" of each isoform for a gene from isoforms.fpkm_tracking file. Does this make sense?

Thank you!

chadn737 · 02-01-2012, 12:23 PM

edgeR and DEGseq take raw counts. They then do their own normalizations. Taking results from cufflinks and trying to use this in any of these programs is not a good approach, even though a lot of people try it for some reason. If you want to use the output of Cufflinks for differential expression, then I would stick to the Cufflinks pipeline and use Cuffdiff.

Otherwise, extract read counts for each gene from your bam/sam/bed file and use this as input for edgeR/DEGseq.

statsteam · 02-01-2012, 01:23 PM

I agree that we'd better stick to cuffdiff for differentially expressed gene analysis. Doe cuffdiff have "paired"-analysis feature for the data with replicates? The paired-analysis feature is the main reason I want to use edgeR.

Thomas Doktor · 02-02-2012, 04:11 AM

Cuffdiff supports replicates but does not handle paired replicates to my knowledge.

Btw, I would recommend using DESeq instead of DEGseq, the spelling is similar but the internal statistical modelling is very different.

dietmar13 · 02-03-2012, 11:51 PM

you could also use SAMseq (samr v2 R-package). this package works with many kinds of designs: paired, quantitative, right censored (like overall survival).
in my hands, SAMseq produced most significant genes (followed by edgeR, baySeq, DESeq, NOIseq, and far far behind cuffdiff) , which were rather robust in bootstrap validations.

my design: 12 normal vs 12 cancer (paired, means from the same patient).

IBseq · 11-15-2013, 12:28 PM

Hello guys,
if anyone knows, could you please tell me why is this happening:

i ran cufflinks on galaxy with default parameters and had satisfactory results.
I then ran the same samples with same parameters except changing max intron length from 300000 to 600000

in the second run have the exact number of transcripts but the FPKM values are much much lower..

Any suggestion?

Thanks,
ibseq

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Getting differentially expressed genes based on RPKM values	casshyr	RNA Sequencing	3	04-30-2012 10:18 AM
DESeq and EdgeR: too many differentially expressed genes!?!?	cutcopy11	Bioinformatics	5	12-08-2011 01:14 AM
Comparing mouse and human differentially expressed genes	stephenhart	General	3	11-16-2011 02:14 AM
Detecting differentially expressed genes using aligner outputs	questioner	Bioinformatics	6	11-03-2011 08:15 AM
gene ontology over-representation of differentially expressed genes	damiankao	Bioinformatics	10	10-20-2011 02:57 PM

02-01-2012, 12:09 PM	#1
statsteam Member Location: Californica Join Date: Sep 2009 Posts: 19	Cufflinks, differentially expressed genes Hi, I am trying to run edgeR or DEGseq using the output from cufflinks. I usually use mapped reads count as an input to edgeR or DEGseq. What cufflinks output do I need to use for an input to edgeR or DEGseq? I am thinking about adding "coverage" of each isoform for a gene from isoforms.fpkm_tracking file. Does this make sense? Thank you!

02-02-2012, 04:11 AM	#4
Thomas Doktor Senior Member Location: University of Southern Denmark (SDU), Denmark Join Date: Apr 2009 Posts: 105	Cuffdiff supports replicates but does not handle paired replicates to my knowledge. Btw, I would recommend using DESeq instead of DEGseq, the spelling is similar but the internal statistical modelling is very different. Last edited by Thomas Doktor; 02-02-2012 at 04:14 AM.

02-03-2012, 11:51 PM	#5
dietmar13 Senior Member Location: Vienna Join Date: Mar 2010 Posts: 107	how many replicates in each condition do you have? you could also use SAMseq (samr v2 R-package). this package works with many kinds of designs: paired, quantitative, right censored (like overall survival). in my hands, SAMseq produced most significant genes (followed by edgeR, baySeq, DESeq, NOIseq, and far far behind cuffdiff) , which were rather robust in bootstrap validations. my design: 12 normal vs 12 cancer (paired, means from the same patient).

02-01-2012, 12:23 PM	#2
chadn737 Senior Member Location: US Join Date: Jan 2009 Posts: 392	edgeR and DEGseq take raw counts. They then do their own normalizations. Taking results from cufflinks and trying to use this in any of these programs is not a good approach, even though a lot of people try it for some reason. If you want to use the output of Cufflinks for differential expression, then I would stick to the Cufflinks pipeline and use Cuffdiff. Otherwise, extract read counts for each gene from your bam/sam/bed file and use this as input for edgeR/DEGseq.

02-01-2012, 01:23 PM	#3
statsteam Member Location: Californica Join Date: Sep 2009 Posts: 19	I agree that we'd better stick to cuffdiff for differentially expressed gene analysis. Doe cuffdiff have "paired"-analysis feature for the data with replicates? The paired-analysis feature is the main reason I want to use edgeR.

11-15-2013, 12:28 PM	#6
IBseq Member Location: uk Join Date: Jul 2012 Posts: 56	Hello guys, if anyone knows, could you please tell me why is this happening: i ran cufflinks on galaxy with default parameters and had satisfactory results. I then ran the same samples with same parameters except changing max intron length from 300000 to 600000 in the second run have the exact number of transcripts but the FPKM values are much much lower.. Any suggestion? Thanks, ibseq