Local Alignment Of Short Reads: Mosaik, Bowtie2, Bwa?

Question

[gaelgarcia05]

Hi everyone! Thanks so much for all your help and advice on this forum.

I am trying to find a good aligner that will help me deal with the problem I'm having. I have a few million reads that span two different (separate) sequences on the genome, the product of an insertion/translocation.

I need a tool that will help me determine the position where the read came from in sequence 1 and the position where it came from in sequence 2, as to determine the new structure in the genome.

I have done some short read alignment so far, and most of the programs I've used require a read to map from end to end to a given contiguous sequence in the reference.

However, I have recently learned that Bowtie2 can now do a local alignment approach, but I am not sure if its soft-trimming capabilities to find the best Local alignment are enough for what I need :

~75 nt of the read belong to sequence 1, and ~70nt (the remaining part of the read) corresponds to sequence 2. However, some nucleotides in the middle of the read may not align anywhere, or may correspond to duplications of sequence 1. So, as you can see, the alignment is pretty messy.

I've come across Mosaik, which I believe also does Smith-Waterman alignment, and in the BWA documentation it is stated that local alignments are performed as a way to rescue those reads that did not align on a first pass.

Do you have any input as to which of these options would be best, or if there are any other options out there that could help?

Thanks! Carmen

Answer 1

[lh3]

I call yours as chimeric alignments. You should use the latest component bwa-mem from bwa. It not only gives you multiple local hits from different parts of a query sequence, but also assigns a confidence score (mapping quality) to each local hit.

As to other tools, bwa-sw can do the same, but its sensitivity and accuracy is lower. I have heard Mosaik can do split alignment, but I have not tried that. For short reads, bowtie2 may also work if you run it in the local alignment mode and let it output multiple hits. However, it is not designed with chimeric alignment in mind. The power will be reduced in some cases. You also need to heavily process its output to get what you want. Yaha is a mapper specifically designed for aligning fusion genes. I have not tried. In addition to yaha, there are other tools to find fusion genes. I have only tried two. Tophat2 makes a lot of false positives with its initial mapping. Star is better but not good enough and it does not give you a confidence value.

[gaelgarcia05]

Thank you, lh3! Chimeric alignments sounds about right! I was unaware that BWA had a new component. I will be sure to try it, then.

I did try TopHat2 with the Fusion option, and it does actually detect those reads that map to both sequences, but only those that have nothing in between, i.e, those that have no unassigned nucleotides to a sequence.

I also heard about Subread m which uses a different algorithm to map reads, but I'm unsure if it will be able to detect a split alignment.

I'll keep things posted. Carmen