Bowtie 2 multiple alignment default mode: how the best one is chosen?

[sbcn]

Dear all,

I am using Bowtie 2 and I have a question about multiple alignments: I am using the default mode which, according to the documentation "search for multiple alignments, report the best one"

What I would like to know is how, if one read aligns to multiple locations with the same alignment score, bowtie 2 chooses to display one and discard the other one?
Are some criteria taken into account or is it in this case "first found first kept"?

I am having doubts about reads mapping into tandemly repeated regions, whether they are assigned or not to the correct location, and how the assignment is done etc. as it is quite important for further SNP calling analysis for example...

Any insight/experience shared will be very helpful.

Thank you!

[sbcn]

Any idea anyone?

[SEQond]

Also interested in this

[xied75]

Haven't read bowtie2's source code, but BWA faces the same issue, i.e. when score equal, how to pick one to report, in BWA, it just call RAND48() to pick one.

I guess all those 'F-M index' aligner work in similar way, including BWA, bowtie2, SOAP2, etc.

Best,

dong