BWA:high amount of unique alignments despite high mismatch tolerance

[moritzhess]

Hi, I am a little puzzled by my data. I aligned 100bp Illumina 3' tags to a reference transcriptome generated from 454 Data using BWA. The Reference has been created using NEwbler, so the reference is organized in groups that contain potetial splice variants originating from one locus . In order to find the best alignment parameters I varied the mismatch parameter (out of curiousity) up to "90" and regarded a hit as unique if all its potential targets would be in one group. I would have expected that the unique alignment ratio would rapidly fall when allowing for a high number of mismatches but this is not observable. Instead the alignment ratio increses slightly and tends to reach a plateau. Has anybody a clue why this happens or am I missing something ?
I attached the change in alignment frequency relative to mismatches as a diagram.
Thanks in advance.

[nilshomer]

If BWA finds a hit for a read with N edits, it will search up to N+1 edits, even if your limit is M (N+1 << M). For example N=2, M=90, it will only search up to 3 mismatches. Of course, this is unless you specify "-N" for the non-iterative mode, which has my favorite command line comment: "slooow".

[moritzhess]

@nilshomer:

Thank's! I have to try that option! (just how to see how fast slooow is). But now the diagrams make more sense.