Too many mismatches aligned by BFAST?

[NGSfan]

Just found an interesting result from one of my BFAST alignments.

I have two replicates of an target enrichment (exome) sequencing run.

In one of the replicates I have pretty clean reads (MQ = 255) and alignments

In the other replicate, I see 12 reads that look out of place (with MQ values ~150). The quality scores per bp are high (~30-38) so it's not sequencing error perhaps, yet these reads don't look like they belong there given the extremely high number of mismatches.

1) How do I prevent/filter out these sorts of alignments? I'm using -a 3 in the BFAST postprocess (uniquely select the best scoring alignment).

2) Also, any one have any theories as to how these reads came about?? What kind of artifact is this? alignment or PCR?

Thanks for any ideas/thoughts/suggestions!

Since there is a 97kb limit for images, I've got my IGV screenshot here:

http://www.sendspace.com/file/r1gg6i

[NGSfan]

the other interesting thing is that the pair mates are in the opposite direction.

Normal pair mates appear as -----> <------
While these highly mismatched reads, the paired mates are in the opposite orientation <------- ------->