miRNA missed by Bowtie alignment

[wieni]

Hi,

I hope this question does not sound to stupid. I have the following problem. I have NGS data from miRNA sequencing. I trimm the reads, so that I have a minimum length of 16nt and a maximum length of 35nt. I used bowtie (not bowtie2) to align/map the reads against a reference. My reference is a bowtie index, based on the human mature miRNAs taken from miRBase. In order to build the index and map the sequences after I exchanged all "U"s by "T"s.

My problem is that I have for example a read with the following sequence:
TCTCCCAACCCTTGTACCAGTGT

Bowtie tells me that this sequence does not match.If I blast it I find it to be mir-150-5p. The identitity is 100%.
My read is 1nt LONGER than the actual reference sequence (TCTCCCAACCCTTGTACCAGTG). Therefore bowtie does not find it. I tried different parameters for bowtie, but only manage to get it, if I shorten my read by 1 nt, which is of course not an option, as this is just a single example we found by random.

Do I make here any principle error? Is it impossible to map a read which is longer than the reference? Any suggestion? Do I need to "artifially" modify my mirBase sequences before I make an index? Like attaching 10Gs at both ends?

[Brian Bushnell]

The best option is to use a different aligner which is able to map reads off the ends of your reference scaffolds. Why are you using bowtie1, anyway? Bowtie2 is much better.

[wieni]

Thanks for the answer.

I am using bowtie1 as it is normally sad to be better when you work with short reads. I already tried bowtie2 with the whole dataset, the results are pretty different and less good. But I did not invest too much time in bowtie2 with regard to miRNA. For other projects I am using bowtie2 as we have long reads normally. I tested the specific sequence I mentioned above with bowtie2 also just "found" it after I shortened it,

Maybe I need to use a completely different aligner?

[cascoamarillo]

You should work with bowtie1 mapping small reads; no doubt. Could you please specify the parameters you are using. And what kind of reference are you using, miRNA database or something similar?
Best

[dpryan]

You would need local, rather than global, alignment for this, which bowtie1 doesn't do (at least if memory serves). Bowtie2 can do local alignment, but you might also follow Brian's advice and just use a different aligner (I suspect that he'd suggest BBMap ).

[Brian Bushnell]

You would need local, rather than global, alignment for this, which bowtie1 doesn't do (at least if memory serves). Bowtie2 can do local alignment, but you might also follow Brian's advice and just use a different aligner (I suspect that he'd suggest BBMap ).

I would have suggested that, but I've never actually tried it with 16bp reads. It works fine with 25bp reads, though. For short reads you can increase sensitivity by using a shorter kmer length and reducing max gap length, using flags like "k=10 maxindel=8". BBMap does global alignment, but it still extends off the ends of scaffolds up to around 44% of the read length by default.

[wieni]

Thanks for the answers!

I was also thinking about the local alignment. Therefore I at least tried bowtie2, but now way.

I will try 2 things:
a) new aligner - as you suggested BBMap, I would also try this SHRiMP
b) modifying my miRNA-reference set

I used bowtie with the following parameters: bowtie -f -l 16 -v 2 -t -a --best --strata