Merging reads mapped on genome and CDS (SOLID data)

[jmandel]

Hi everyone,


We are interested in discovering fusion genes in RNA-seq data from a
certain tumoral type, and FusionSeq seems to be a tool of choice for
this particular topic. Our RNA-seq data comes from the SOLID sequencing
platform, it is paired-end with 50bp-long reads on the 3' side and 25
bp-long reads on the 5' side.

Our main concern is the choice of the mapping strategy. We use Bfast+BWA, that has the advantage of being sensitive and capable of handling very short reads. Here is our mapping strategy : we aligned reads on the reference genome, then we extracted the non-aligned reads, and we aligned them on a library of all known human CDS. We then want to use FusionSeq for the search of fusion genes. As far as we understood FusionSeq takes as an input reads aligned on a genomic reference under the MRF format. Our concern here is the following : how to merge these two mappings (genome and CDS) into a single one, that we can then feed to FusionSeq ? Do we need to convert the coordinates of reads aligned on the CDS to genomic coordinates ? And if so, how ?

Thank you by advance for your help.

Jonas and Jaydutt

[jjohnson]

Jonas and Jaydut,

We use SOLiD extensively here at EdgeBio and finds that when aligning transcriptome data, even more than genomic data, the alignments really matter. Just aligning with bfast or bowtie without taking into account the complexities of transcriptome data won't yield optimal results. I suggest Tophat or novoalign (see below) but we find Tophat doesn't map solid data extremely well due to its error profiles. It is more tuned to Illumina data.

We have consolidated around novoalign for our alignments for genomic and transcriptomic data, and novoalign has recently released a recipe to align your RNASeq reads to the whole genome + splice junctions. It shows you how to convert to genomic coordinates as you ask, and leverages trusted R packages such as Useq and DESeq. The result will be sorted alignments in BAM format against the transcriptome, genome and splice junctions. How this is fed into fusion detection software should be a trivial conversion. We are in the process of testing this with an internal project of 32 samples.

If you use the recipe, I would be interested in comparing notes.