BWA & multiple alignments per read

[Giles]

Bowtie clearly indicates in its manual that it will randomly assign any read for which there are multiple equally valid alignments. I can not see any such discussion regarding the "mem" aligner for BWA. Can anyone tell me how BWA handles multiple alignments per read, and what the default settings are?

[lh3]

Only outputs the best hit or a random one if there are multiple. You can ask to report more with -a. Note that bwa-mem may report chimeric alignments, but those are different from multiple mappings you are talking about.

[A_Morozov]

Can anyone please explain me why the default behaviour is to randomly choose one of mappings? Making sense of multiple mappings is big enough headache without mapper hiding half the results from you.

[swbarnes2]

Can anyone please explain me why the default behaviour is to randomly choose one of mappings? Making sense of multiple mappings is big enough headache without mapper hiding half the results from you.

I guess it's so that coverage across repetitive regions is comparable to that of unique regions. If you give all the reads to each of 10 repetitive regions, coverage at each will be 10x higher than the rest of the genome.

If a region is repeated 100 times, it will have 100x coverage as compared to the rest of the genome, which would really screw with trying to count how well the rest of the genome is covered.