Alignment at exon-exon junctions

[Boel]

Hi All,

I have RNA seq data (human, paired end, 72 bases, Solexa) from a bunch of samples. In one of my analyzes I have realized that a big problem is the fact that reads get mis-aligned at exon exon junctions, see attached illustration. So far I have mainly been using Bowtie, and the "errors" I am seeing makes sense given how Bowtie works. So reads will map with a small "overhang" at the exon border, especially if there is some seq. similarity at that border and on the other side of the junction (the other exon). This leads to that I will get what looks like sequence variation, in the attached illustration for example the red AA on the left hand side will look a SNP or an editing event.

One easy approach to "correct" for this would be to only analyze data that falls within exon borders, but this will cause a loss of information. So my question is: Are there other aligners that are better at aligning around exon junctions?

Another option is to align reads to the transcriptome instead, which I have been doing, but genome annotation then gets lost which is a major problem.

Thankful for any thoughts about aligners that would work well in this situation!

Boel

[Thomas Doktor]

First of all, I'm not sure if you have used Bowtie itself or TopHat, which uses Bowtie as it's aligner?

In any case, I don't think you can completely eliminate the problem with misaligning reads right at the intron-exon border, because you need a certain minimum length of coverage on either side of the intron in order to map a read across an exon junction. What I would do, is filter out all variations found within a distance of, say 5 bases, from the exons. This will minimize the number or true positives that you reject, especially considering that this region is already particular sensitive to mutations and seldom have neutral SNPs within them. If you are looking for disease causing mutations which for instance cause intron retention or usage of alternative splice sites, you should be able to detect this and then change the filter for those particular cases.

Another RNA-seq aligner you might try using is MapSplice which, in my experience, does a little better job than TopHat/Bowtie.

[Boel]

@ Thomas Doktor:
I'm using Bowtie itself for this particular analysis.
One could imagine that there could be an aligner that after aligning reads and defining exon-exon junctions actually re-aligns the reads that cross that border. Could be a neat approach. Maybe something I should develop myself.