[Brian Bushnell]

I suggest you try BBMap - it can map RNA-seq data with very high error rates, including indels, and in my testing greatly outperforms GSNAP. For RNA-seq you have to include a few extra flags from default, as mentioned in that thread.

BBMap as compiled will handle reads up to 500bp. I don't really know how long Ion Torrent reads are so if that's too short, tell me and I can compile it with support for longer reads.

[GenoMax]

BBMap as compiled will handle reads up to 500bp. I don't really know how long Ion Torrent reads are so if that's too short, tell me and I can compile it with support for longer reads.

Torrent reads are well below 500 bp (more like ~100- 150 bp) so this version should be ok.

[mbzmg1]

I suggest you try BBMap - it can map RNA-seq data with very high error rates, including indels, and in my testing greatly outperforms GSNAP. For RNA-seq you have to include a few extra flags from default, as mentioned in that thread.

BBMap as compiled will handle reads up to 500bp. I don't really know how long Ion Torrent reads are so if that's too short, tell me and I can compile it with support for longer reads.

Have you tested it against GSNAP with Ion Torrent data?
I will take a look and probably test it out if mismatch etc can be set as a percentage/fraction of read length. The problem with Tophat2/Bowtie2 and Ion Torrent data is a bias against longer reads which have more errors than shorter reads. This is also a slight problem with GSNAP as it will only allow a single indel. It would be great if BBMap allows for tweaking these parameters.

[IonTom]

Currently i am using NextGenMap for Alignment.
This works very well and has been adapted for varying read length and
the homopolymer error.

To make it behave closer to tmap you can try to increase --gap-read-penalty and --gap-ref-penalty.


I have not tried MOSAIK yet, which could also work well.

[Brian Bushnell]

Have you tested it against GSNAP with Ion Torrent data?
I will take a look and probably test it out if mismatch etc can be set as a percentage/fraction of read length. The problem with Tophat2/Bowtie2 and Ion Torrent data is a bias against longer reads which have more errors than shorter reads. This is also a slight problem with GSNAP as it will only allow a single indel. It would be great if BBMap allows for tweaking these parameters.

I've tested it against GSNAP with synthetic Illumina data containing many indels and mismatches per read. BBMap has no limit to the number of indels or mismatches, but if there are too many the mapping score will drop below a cutoff and be rejected. A 150bp read can map with up to roughly 13 single-base-pair indels, or 29 snps, or a single deletion of up to (I think) 2.8 megabases, or some combination of things like that, while staying above the score cutoff. You do have to change the maxindel flag for RNA-seq, though; by default it is 16000; for vertebrates you should change it to around 200000 or so. Anyway, it can align reads just fine that spans 2 or 3 or even more introns.

BBMap is not biased against longer reads, because the scoring is always considered as a function of the read length; so a 100bp read with 1 mismatch would have the same score as a 200bp read with 2 mismatches. After factoring in ambiguity, the 200bp read might end up with a higher score, because a shorter read is more like to have multiple possible mapping locations, which incurs a score penalty.

Here's a comparison of BBMap and GSNAP:

https://drive.google.com/file/d/0B3llHR93L14wbks0ZURFcFhFR1E/edit?usp=sharing