RNA-seq pipeline on Ion-Proton data

[super0925]

Hi I am a rookie bioinformatics who will process the rna-seq from Ion-proton(Single-ended).
The pipeline for processing which I know until now is
Bowtie, Tophat, Cufflinks, and cummeRbund for RNA-seq alignment, assembly, DE analysis, and .visualization.
Any other options?
Or What is this advantage or disadvantage of this pipeline?
Thank you!

[BrianJames]

There is a link a the Ion Community that describes a pipeline. You have to register, but it is free.

http://ioncommunity.lifetechnologies.com/docs/DOC-7062

It uses a combination of Bowtie2 and Tophat2


I added a step to filter reads with Bowtie2. I used a filter file that I used to use with SOLiD data (ribosomes, adapters, tRNAs and such).

Brian

[super0925]

Thank you! I got it.
But this pipeline which is suitable for Proton data is only reach the mapping and assemble. So after that steps which pipeline is more suitable? I mean for the differential analysis?
Cufflinks or Deseq or EdgeR? Do you have any recommendation ?
Cheers

[super0925]

There is a link a the Ion Community that describes a pipeline. You have to register, but it is free.

http://ioncommunity.lifetechnologies.com/docs/DOC-7062

It uses a combination of Bowtie2 and Tophat2


I added a step to filter reads with Bowtie2. I used a filter file that I used to use with SOLiD data (ribosomes, adapters, tRNAs and such).

Brian

Thank you! I got it.
But this pipeline which is suitable for Proton data is only reach the mapping and assemble. So after that steps I got the bam file, then which pipeline is more suitable? I mean for the differential analysis?
Cufflinks or Deseq or EdgeR? Do you have any recommendation ?
Cheers

[BrianJames]

I use Cuffdiff in cufflinks2. It's pretty straight forward, but I can see if I can find the command I ran on Tuesday, if you need it.
Brian

[super0925]

I use Cuffdiff in cufflinks2. It's pretty straight forward, but I can see if I can find the command I ran on Tuesday, if you need it.
Brian

Thank you Brian! Cuffdiff is really good but so far I don't know hot to compare the achievement between different methods (i.e. which is more suitable for the Proton data)

[BrianJames]

You might want to take a look at this thread.

http://seqanswers.com/forums/showthread.php?t=17678

I know I will look at it more when I get the chance. I must admit that I like Cuffdiff because since I've already installed cufflinks2 it is a simple linux command. To address your question (not answer it), once you have your read data aligned, I don't think the Proton platform is going to impact performance of differential expression options. I would like to work with the best package out there, but they all work pretty good, and are mostly impact statistical significance calculations. INMHO, good starting material (RNA quality), a well designed assay, and good biological replicates will have a much bigger impact on the quality of the eventual results.

Brian

[super0925]

You might want to take a look at this thread.

http://seqanswers.com/forums/showthread.php?t=17678

I know I will look at it more when I get the chance. I must admit that I like Cuffdiff because since I've already installed cufflinks2 it is a simple linux command. To address your question (not answer it), once you have your read data aligned, I don't think the Proton platform is going to impact performance of differential expression options. I would like to work with the best package out there, but they all work pretty good, and are mostly impact statistical significance calculations. INMHO, good starting material (RNA quality), a well designed assay, and good biological replicates will have a much bigger impact on the quality of the eventual results.

Brian

Thank you Brian! So nice you are!

[jay]

For DE analysis I would definitely recommend using multiple methods, e.g. cuffdiff, EdgeR, etc. BitSeq looks nice and I heard a talk about CASPAR which looks nice. They each use a different model to approximate the distribution of read counts, so they each give a different answer. It is worth trying several and taking the union/intersection.

[mbzmg1]

Our initial attempts at using Tophat2/Bowtie2 to map Ion Torrent RNAseq data were pretty disappointing (~30% aligned) and turned out to be biased against alignment of the longer reads (see attached image). Our average read length was ~75 bases, and longest ~150+ bases. This bias may be due to fixed penalties applied for gaps/mismatches per read rather than as a fraction of read length.

We had much better success with GSNAP (~70% alignment) and have used HTseq-count and DESeq for our differential expression analysis. Gene lengths and RPKMs can easily be calculated separately if required.

It is also worth noting that the second stage of the protocol at http://ioncommunity.lifetechnologies.com/docs/DOC-7062 will not add back any reads that should align over exon junctions. One might therefore get an underestimate of these positions.

[Brian Bushnell]

BBMap should be ideal for mapping Ion RNA-seq data. In my testing, it vastly outperforms GSNAP in all cases, and produces Tuxedo-compatible output.

If your reads are all shorter than 500bp, you can run it in normal mode; otherwise, you can run it in PacBio mode (with the script mapPacBio8k.sh).

[BrianJames]

BBMap should be ideal for mapping Ion RNA-seq data. In my testing, it vastly outperforms GSNAP in all cases, and produces Tuxedo-compatible output.

If your reads are all shorter than 500bp, you can run it in normal mode; otherwise, you can run it in PacBio mode (with the script mapPacBio8k.sh).

Hi Brian,
I think this is the link you want http://seqanswers.com/forums/showthr...568#post133568

Brian

[Brian Bushnell]

Thanks, that's exactly what I wanted!