RNA-Seq is very efficient to estimate gene expression level in terms of RPKM/FPKM and to perform differential expression analysis then. Which of the following methods do you prefer to calculate the RPKM/FPKM value/perform DEG analysis especially when a single gene has multiple isoforms?
RPKM calculation
a). Reads falling on any exon of the gene are counted and then normalized by the total length of all exons and library size;
b). Only reads falling on the constitutive exons are counted and normalized by the length of constitutive exons and library size;
c). Reads falling on the epresentative isoform (eg. longest isoform) of a gene are counted and normalized by isoform length and library size
d). by ERANGE software (http://www.nature.com/nmeth/journal/v5/n7/full/nmeth.1226.html)
e). by cufflinks/cuffdiff software, which made effort to locate reads in isoform/transcrip level.
Differential Expression Gene Analysis
a) DEGseq, b) EdgeR, c), Cuffdiff
By the way, what's your comments on Cufflinks/Cuffidff software? Actually, I heard of both positive and critical opinions, which makes me confused on whether it is reliable.
Dear Chen,
I would recommend you to follow this post "https://www.researchgate.net/post/What_is_your_favorite_DEG_test_for_RNA-seq_data" . I hope you will get your answer here.
And if you are not able to access it then please search with the question "What is your favorite DEG test for RNA-seq data?". You will definitely get it.
Thanks, Singh, I found the post, and it is very helpful.
Thanks, Yang. and any explanation why you prefer DESeq2 rather than other method? have you ever performed comparasion between different methods?
Hello Jia-Yu. I've tried the last version of cufflinks/cuffdiff for RNAseq isoforms expression quantification and differential expression on a set of two samples with 2 replicates each, and performed these analyses 3 times. I have now 3 different results. In my opinion until this problem is not fixed I suggest you to use DESeq2.
Thanks Valerio!
Do you mean the results are different even when you performed the analysis with the same parameters?
Why does the previous version of cuffdiff have a bad reputation?
Someone also evaluated the updated cuffdiff and found that it was not so bad. "But cuffdiff v 2.1.1 was just released and it was significantly changed over previous versions. I did a quick comparison with results from edgeR, of 414 DE genes claimed by edgeR at FDR 0.05, 302 of them overlap DEGs (total 477 at FDR 0.05) from cuffdiff . "
Hi All,
I think the software/methodology to choose depends on what you want to do. I am only familiar with RNA-seq in non-model plants, where gene expression can be 'straightforwardly' carried out with Expectation-Maximization (RSEM) software (http://deweylab.biostat.wisc.edu/rsem/README.html) bundled with Trinity package and Differential gene expression with edgeR (http://www.bioconductor.org/packages/2.11/bioc/html/edgeR.html) software package to examine differential expression of replicated count. This will allow you to do clustering and to draw plotMDS for samples to observe relationship. To me, edgR bio-conductor package worked well and allowed me to narrowed down to a handful of FG.
Yes Jia-yu, using the same parameters led to different results. This is odd for me. So I suggest to use DESeq2, but also other methods such as RSEM, ISOem and so on. They work on different principles for the deconvolution of expression levels and DEG estimation. Rohan Chaubal gives you a good advice, try and confirm by qPCR is better if you can perform such experiments.
So what I guess you are asking is, given this workflow, which method would be best to use.
DESeq and edgeR are pretty much equivalent with DESeq possibly being a bit more conservative.
Cufflinks/cuffdiff are for differential exon usage analysis (different from differential gene analysis for edgeR/DESeq).
DESeq2 is the new version of DESeq. DEXseq is another program for differential exon usage
Some things to consider:
- How are you defining your exons? Often times for humans knownGenes track from UCSC is used but ensembl genes track or gencode annotations could also be used.
- How do you dealing with ambiguous read count overlaps? (see http://www-huber.embl.de/users/anders/HTSeq/doc/count.html)
- Are you filtering non-expressed genes?
If you are working w/novel organism, you might want to try out Trinity:
http://trinityrnaseq.sourceforge.net/
If you are very interested in differential exon usage, you might want to try out Tuxedo tools (Bowtie, Tophat, Cufflinks, cummeRbund)
We carried out a similar experiment using DESEQ for differential Gene analysis and found some interesting stuff which helped us to dig deeper into Differential Exon usage for which we used DEXSEQ...
edgeR and DESEQ are fairly closer in the type of analysis you do and Its definitely your choice to decide..
DEXSEQ is fantastic when it comes to Differential Exon usage analysis... Also you might want to make sure if you are using the right reference genomes and check for the right number of gene/exon counts.. That will be your next crucial step in your analysis ...
Thanks, Rohan. We do hope that this paper will be very useful for RNA-Seq analysis. We sent the preliminary version of the paper to authors of all packages that we evaluated and incorporated some of their comments in the final version. Thanks for the suggestion to start a thread on seqanswers. I will pass it to my colleagues.
Just came across this:
http://is.landesbioscience.com/journals/cib/article/25849/
Note that if you decide to use RSEM for read alignment, then you might want to consider using EBSeq to test for differential expression.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics
- Bioinformatic Tools