bwa mem: Get all possible alignments for multi-mapping reads

[nk]

When a read maps equally well to multiple locations in the genome, `bwa mem` seems to set the mapq to 0 and report one at random. So if there are two possible locations, I may end up with ~50% of my reads on chr1 and ~50% of my reads on chr2. However, I only care about the region on chr1, so I currently lose 50% of my reads because they end up getting placed on chr2.

Is there any way I can get bwa to report multi-mapping reads in both locations, so that I get 100% of the reads on chr1 (as well as on chr2)?

Or would I manually have to resolve the XA tag to generate these additional alignments? Has anybody done this before?

[GenoMax]

Not answering your question directly. This would be easy with bbmap.sh from BBMap suite. You would choose ambig=all option to get all possible alignments for multi-mappers.

[dcameron]

If you need to use bwa mem, you'll want to use the -h and/or the -a command-line options but I don't think that'll do exactly what you're after. A number of other aligners can report in the manner you have requested (eg bowtie2, novoalign).

Be aware that some reads can have thousands of mapping locations so your raw output file can be over an order of magnitude larger than the standard output reporting only the best alignment.