Very short reads (20-22) , bowtie and tophat not doing gapped alignment

Question

[manekineko]

I have a reads which are quite short (~22nt), and I want to find which mapped to intron-exon boundaries.

I have run tophat with segment=10 but did not find any spliced reads - probably impossible, so something must be wrong with the options I'm using?

Also tried bowtie2 --local, the option I do not see any clipped aligned reads (at least I cannot see in the IGV, is there a way to count the number of trimmed reads that are aligned in --local mode if any?)

[Brian Bushnell]

20bp reads are too short for useful spliced alignments unless the introns are extremely short, on the order of a few dozen bp. That said, BBMap can map such reads spliced, with a command like this:

bbmap.sh in=reads.fq ref=ref.fa out=mapped.sam maxindel=100 k=10 slow

k=10 and slow are optional but recommended for mapping such short reads spliced. You can also add the "local" flag but that may reduce the number of reads spanning an intron.

[Istvan Albert]

Try reducing the seed length, the read has to be longer than the seed length for the alignment to work. Default seed lengths are around 20.

[manekineko]

you have no option for seed in tophat or am I wrong? this option is the segment one and I already decrease it to 10

[Istvan Albert]

bowtie is the aligner that users the seed lengths - it takes seed length via -L if you use it via tophat the option is called --b2-L I think.

[karl.stamm]

When they speak of spliced alignments they span across an intron connecting two exons. If you want an intro. Exon boundary that's a genomic aligner without splicing.

Answer 1

[mark.ziemann]

Hi manekineko, I've done simulations with 21 nt hairpin derived reads for rice and human microRNAs and found that even with 1bp indel, the reads are either unmapped or a large proportion of spurious mappings with 16 different mappers. Because performance was somewhat better for the smaller genome, you could get better results by mapping directly to a library of exon junctions.

In addition to Brian's suggestion of BBmap, I would recommend trying

• Bowtie2 with -very-sensitive-local parameter
• Bowtie2 with -very-sensitive parameter
• Bowtie1 --best --strata parameter
• SMALT with k=10 s=1 parameters