Isoform expression quanification from rna seq: flood of tools

[Maayanster]

I'm trying to figure out which expression quantification program to integrate into the RNA analysis pipeline at my centre.

Since the review by Garber et al. in Nature Methods (June 2011), and the review by Martin and Wang in Nature Reviews (October 2011), there has been a veritable flood of new tools, particularly in the category of transcriptome reconstruction and isoform expression quantification.

From reading those reviews and a quick search through pubmed, I compiled the following list of programs that are capable of giving transcript-level expression information as of December 2012:


Scripture
Cufflinks
MMSEQ
Alexa-Seq
MISO
IsoLasso
CEM
cuffdiff 2
RSEM
FDM
IsoformEx
NSMAP
IQSEQ
DEXSeq
DSGSeq
IsoEM
IsoInfer
iReckon
BitSeq
eXpress
SLIDE
DiffSplice
SplicingCompass
SailFish

This doesn't include tools like DESeq, EdgeR and baySeq that supply only gene-level differential expression.

I've started reading some of the papers for these tools, but honestly, without that much background in statistics, their relative advantages/disadvantages are pretty opaque. There's discussion in these papers of approaches to normalize various biases, handling biological replicates, handling longer read sizes, and providing information on gene- transcript- and in some cases exon-levels.

Anyone have advice or comments about any/all of these new choices?

UPDATE: see below for posts with my impressions of the tools that I've chosen to test out. Learning a lot! All the summaries of tools I've tried are now in one post:

http://seqanswers.com/forums/showpost.php?p=102911&postcount=60

[turnersd]

I don't have much in the way of advice, but I recently wrote a synopsis of the isoform-resolution differential expression problem in response to the cuffdiff 2 paper.

http://gettinggeneticsdone.blogspot.com/2012/12/differential-isoform-expression-cuffdiff2.html

[EduEyras]

I'm trying to write a short review on these and other tools. Not explaining the details of the methodology, but actually trying to pinpoint other aspects that distinguish them. The aim is to help out people having an analysis in mind and wanting to choose the right tool. I'll be happy to share as soon as I have a draft ready. I can already tell you some issues are not clear even after reading the publications.

[turnersd]

I'm trying to write a short review on these and other tools. Not explaining the details of the methodology, but actually trying to pinpoint other aspects that distinguish them. The aim is to help out people having an analysis in mind and wanting to choose the right tool. I'll be happy to share as soon as I have a draft ready. I can already tell you some issues are not clear even after reading the publications.

Definitely share this here once it's published.

[EduEyras]

Hi,

I've put here http://regulatorygenomics.upf.edu/So..._and_splicing/

a preliminary list of genome-based methods for isoform reconstruction/quantification and for
differential splicing (isoform/event levels)

I've separated them according to whether they need a gene annotation or not, and also
if they use a transcriptome as input (transcript sequences from a known annotation).
I think this could be already an useful separation, as the tool to be used will depend
on things like: how much annotation is available? Do you want to predict novel isoforms or just quantify known ones? etc.

I haven't added yet the transcript assembly methods (off-genome), but they also generally provide a quantification.

Some methods can do various things, like SOLAS or MISO. Some methods are actually part of
a set of tools, like Cufflinks, Cuffdiff2 and RABT, which are part of Cufflinks.

Any comments or suggestions are welcome. I'll keep on updating it with extra info

Eduardo

[turnersd]

Very useful - thanks for posting.

[krespim]

Nice list.

MISO

As far as I know, and I have been using MISO regularly, it does not give information on isoforms. It is very much "exon-centric".

I've started reading some of the papers for these tools, but honestly, without that much background in statistics, their relative advantages/disadvantages are pretty opaque.
(..)

Anyone have advice or comments about any/all of these new choices?

I also feel the same. When I choose a tool I always look for information on validation rates, that is, where the predictions reproduced at the experimental level? It really does not matter if uses Baeysian inference or binomial distribution if the predictions are not validated in the "real data". I also take into account easy of use (very often compiling the tools is a nightmare), and whether the output understandable.

I know these practical/trivial considerations but IMO they are worth consideration.

[jparsons]

I have data that I believe would be good for comparing these tools, as it has a built-in way to validate DE calls at either gene or isoform level. However, I don't have computational infrastructure or skills built up to analyze it quickly/properly - I've spent days trying to get MISO and RSEM to compile and accept the inputs with no successes.

If there were, for example, an (accessible) galaxy instance somewhere with many or all of these tools pre set-up, I would be overjoyed.

[Maayanster]

Hi,

I've put here http://regulatorygenomics.upf.edu/So..._and_splicing/

a preliminary list of genome-based methods for isoform reconstruction/quantification and for
differential splicing (isoform/event levels)

wow, thanks

[turnersd]

I have data that I believe would be good for comparing these tools, as it has a built-in way to validate DE calls at either gene or isoform level. However, I don't have computational infrastructure or skills built up to analyze it quickly/properly - I've spent days trying to get MISO and RSEM to compile and accept the inputs with no successes.

If there were, for example, an (accessible) galaxy instance somewhere with many or all of these tools pre set-up, I would be overjoyed.

Want to share some of this data with the community? http://figshare.com/

[jparsons]

Want to share some of this data with the community? http://figshare.com/

It will be released relatively soon. For now, though, I can't.

[Maayanster]

If there were, for example, an (accessible) galaxy instance somewhere with many or all of these tools pre set-up, I would be overjoyed.

There seems to be just that: a cloud platform for rna seq analysis, which is actually on the galaxy server called oqtans. It has a choice of tools (though not the most recent ones, and disproportionately ones developed at the Ratsch lab)

http://www.raetschlab.org/suppl/oqtans

http://galaxy.raetschlab.org/

has anyone used this?

[jparsons]

There seems to be just that: a cloud platform for rna seq analysis, which is actually on the galaxy server called oqtans. It has a choice of tools (though not the most recent ones, and disproportionately ones developed at the Ratsch lab)

http://www.raetschlab.org/suppl/oqtans

http://galaxy.raetschlab.org/

has anyone used this?

Unfortunately, of the 20 tools on the list on the start of the page, only cufflinks/cuffdiff are on this server. rQuant/rDiff are potentially candidates, but 1 tool against the 'default' does not a thorough comparison make.

[Maayanster]

Giving this a bump in case more people have some input.

It generally seems that not enough people have experimented with the various options yet. I'm starting to install a whole bunch of these, and I'll try to update as I discover things.
I've updated the list with eXpress, which was just published in Nature Methods, and SLIDE which is older.

also, there's a poll on the RNA seq blog http://www.rna-seqblog.com/

[bruce01]

Just to mention SpliceGrapher (http://splicegrapher.sourceforge.net/), it gives some really nice plots. EDIT: just seen it is up on the list linked to above.

On the side of what works best I wrote my own script to output information I thought would be of use in determining differential isoforms between conditions. Prior to this I tried about 6 of those listed and honestly could not dig through the piles of output to decide between them. I am working on tissue though and so differences may be less obvious.