comparing results by cuffdiff, edgeR, DESeq

[PFS]

Has anyone tried to compare the results from the various tools that offers differential expression analyses for RNASEQ data?

I understand they have different underlying models and assumptions, but I would expect some overlap. At a first glance, when I tried to compare the results on my data, I got completely different DE genesets and I am puzzled.

[chenyao]

Has anyone tried to compare the results from the various tools that offers differential expression analyses for RNASEQ data?

I understand they have different underlying models and assumptions, but I would expect some overlap. At a first glance, when I tried to compare the results on my data, I got completely different DE genesets and I am puzzled.

This is quite interesting question?

I also want to know the answer. With my experence on microarray, different statistcal method may generate different DEG list, but they should have a lot overlap.

[mbblack]

Has anyone tried to compare the results from the various tools that offers differential expression analyses for RNASEQ data?

I understand they have different underlying models and assumptions, but I would expect some overlap. At a first glance, when I tried to compare the results on my data, I got completely different DE genesets and I am puzzled.

I analyzed the same RNAseq data set using cuffdiff, DESeq and EdgeR. The data was a whole transcriptome SOLiD 4 fragment run for 3 control and 3 treatment animals (mice), so the final mapped reads ranged from approx. 71M reads, to a high of approx. 171M reads (because of the layout of the barcoding on the slides they were not all equivalent). Raw counts were extracted from BioScope mapped reads using BAMtools and a UCSC RefGene bed file as reference.

Going solely by FDR < 0.05 as a cutoff:

EdgeR - only 217 significantly differentially expressed genes
DESeq - 337 significantly differentially expressed genes
There were 198 genes in common in those two lists.

Cuffdiff gave 202 significantly differentially expressed genes by q-value, but I don't know offhand how many of those are genes common to the other two.

The issue that has me really tearing my hair out is that I have microarray (affy) data for these same animals. A LIMMA analysis of those 6 arrays gives some 3000 significant genes by FDR < 0.05. With the coverage I have with the RNAseq data, I should at least be comparable with the array data, not left with a difference in significant genes of 1000s.

We were expecting to approach or exceed array results with about one tenth of the mapped reads I have for these initial runs.

[songyj]

how wonderful

I analyzed the same RNAseq data set using cuffdiff, DESeq and EdgeR. The data was a whole transcriptome SOLiD 4 fragment run for 3 control and 3 treatment animals (mice), so the final mapped reads ranged from approx. 71M reads, to a high of approx. 171M reads (because of the layout of the barcoding on the slides they were not all equivalent). Raw counts were extracted from BioScope mapped reads using BAMtools and a UCSC RefGene bed file as reference.

Going solely by FDR < 0.05 as a cutoff:

EdgeR - only 217 significantly differentially expressed genes
DESeq - 337 significantly differentially expressed genes
There were 198 genes in common in those two lists.

Cuffdiff gave 202 significantly differentially expressed genes by q-value, but I don't know offhand how many of those are genes common to the other two.

The issue that has me really tearing my hair out is that I have microarray (affy) data for these same animals. A LIMMA analysis of those 6 arrays gives some 3000 significant genes by FDR < 0.05. With the coverage I have with the RNAseq data, I should at least be comparable with the array data, not left with a difference in significant genes of 1000s.

We were expecting to approach or exceed array results with about one tenth of the mapped reads I have for these initial runs.

Wow..
I only used Tophat/cufflinks ,2 ,only 2 significantly differentially expressed genes in isoform_exp.diff
seems like maybe I could try DESeq, hope it will give me more differentially expressed genes

[super0925]

Hi All I am a rookie in RNA-seq.
I found a problem that gene ID output from Cuffdiff and egdeR/DESeq are different so that I cannot find the common DE genes.
I use the data set with the Drosophila melanogaster genome.
The edgeR output gene ID (e.g. FBgn0000370,FBgn0000500,…)
But the Cuffdiff output gene ID is something like (XLOC_000028,XLOC_000038,…) and gene symbol (e.g.,KH1,RpLP1,…).
Could you please recommend any software or command to translate them automatically? Could you have command line for me to use to uniform or join them?

[Thias]

BioMart is an option. I frequently use it as R package ("biomaRt"), but there is also an easy to use online tool.