Which Program Should I Test To Map Rnaseq Data? Did Anybody Tried Gsnap?

Question

[Geparada]

Currently I'm working with BLAT alignment data downloaded from UCSC browser and own generated alignment data using GMAP.

In the next few weeks I want to give the big next steep to work with RNAseq row data. I'm basically interested to study the splice junction (canonical and non-canonical).

I want to test the memory-eficents and fast RNAseq alignment programs. My candidates are:

• Bowtie
• HMMSplicer
• GSNAP
• BLAT (for the 100 bp reads RNAseq data)

I read the GSNAP paper and it's looks very promising, but I didn't found commentaries about it. Anybody tried it?

Anyway, What others programs do you advice to test?

Thanks for your time!

[Malachi Griffith]

A list of alternative splicing tools and resources can be found in this post: Recommended tools for alternative splicing detection from RNA-seq data

Answer 1

[Darked89]

You may also check GEMlibrary. There is an effort to compare spliced mappers called RGASP, which recently released the results of the round 2.

It is hard to make head or tail from bunch of graphs, but GEM was quite high on the list. BTW, Tophat did not enter this contest, and results presented were often obtained using versions not available to the public or possibly extra tuned for the RGASP.

For a list of various RNA-Seq mappers you may check this page (shameless self plug).

[Leszek]

I vote for GEM as well. I've tried BLAT, Bowtie and GEM. In my opinion GEM is the best!

[Varun Gupta]

Hi Leszek

With bowtie did you ever get cigar string with N's. I don't ever get it so i guess bowtie alone by itself is not a splice align aware and hence not a good candidate for rna seq data Please let me know

Varun

[Ashutosh Pandey]

Tophat is a splice align aware from the Bowtie group.

[lh3]

+1. Quite interesting. Nonetheless, I am usually skeptical about anything not published. RGASP is potentially valuable, but not benchmarking TopHat and a few other popular mappers makes it less useful.

[Geparada]

I don't understand how to read the RGASP results, can you give me a clue?

[Darked89]

@Geparada: To be honest I had a chance to see and discuss the presentations from the last day of RGASP meeting in Barcelona. Even then the evaluation was far from obvious, as i.e. some groups submitted multiple predictions for the same data set. What I got was that save for some not as yet available mapper which was fine tuned using existing annotation, GEM (submitter Tyler Alito) was fairly consistently high on the list, together with unreleased STAR software by Alexander Dobin (no idea under which name this was submitted). I will try to contact ppl publishing RGASP results.

Answer 2

[Zheng Chen]

If I understand your question correctly, you are trying to study the splice junctions using the alignment results from RNA-Seq.

For my personal experience, I vote for Tophat. In general, TopHat generates database of possible splice junction sites with and without the gene model annotation. It utilizes bowtie to perform two-stage of alignment. Firstly it maps short reads that can be aligned against the genome, and in the second stage, it aligns unmapped reads back to potential splice junction sites generated in the first stage.

[brentp]

@Geparada, the dinucleotide thing depends on your read-length. From the current manual: "With long (>=75bp) reads, "GT-AG", "GC-AG" and "AT-AC" introns be found ab initio. With shorter reads, TopHat only reports alignments across "GT-AG" introns"

[Geparada]

I'm very interesting in study introns with non-canonical splice juntions and as far I know TopHat search splice juntions based on canonical terminal dinucleotides (GT-AG).

"TopHat (Trapnell et al., 2009), works by first creating exonic coverage islands from short-reads and then, based on canonical intron terminal dinucleotides (ITDN), which exist in these islands, identifies possible splices between neighboring exons. Like QPALMA, TopHat is strongly built around the idea of canonical ITDN. "

[Geparada]

I'm very interesting in study introns with non-canonical splice juntions and as far I know TopHat search splice juntions based on canonical terminal dinucleotides (GT-AG).

I read in a paper: "TopHat (Trapnell et al., 2009), works by first creating exonic coverage islands from short-reads and then, based on canonical intron terminal dinucleotides (ITDN), which exist in these islands, identifies possible splices between neighboring exons. Like QPALMA, TopHat is strongly built around the idea of canonical ITDN. "

[Geparada]

@brentp, I need a program that find introns no matter what dinucliotide it has.

Answer 3

[Jeremy Leipzig]

RUM is a very sensitive tool these types of splicing studies:

RUM is an alignment pipeline that maps reads in three phases. First it maps against the genome using Bowtie, then it maps against a transcriptome database using Bowtie, then it maps against the genome using Blat. The information from the three mappings is merged into one mapping. This leverages the advantages of both genome and transcriptome mapping as well as combining the speed of Bowtie with the sensitivity and flexibility of Blat.

[Geparada]

I'll check that pipeline. Currently we are using a very similar pipeline starting with Bowtie aligments too, but I'm not sure if BLAT is the best option for us, because BLAT take a very long time to map short reads.

[Jeremy Leipzig]

I think RUM does some triage to BLAT only the ones that might be spliced

[Mikael Huss]

I've had good experiences with RUM, it didn't take that long to run because only a subset is done with BLAT.