BWA Reports High Mapping Quality for Poorly Mapped Pairs

[mdjones66]

I realized after I, serendipitously, ran "bwa sampe" on the same identical alignments (i.e. sai files from the same fastq) that the second mate in the pair would almost always have many mismatches. The surprising thing is that a second paired read with as many as 79 out of 101 mismatches could have a high mapping quality.

I am sure this has something to do with the Smith Waterman alignment that is attempted for the second read.

See: http://seqanswers.com/forums/showthread.php?t=14665

I have a few questions:

1) Why does bwa try and rescue reads that don't map well with bwa aln?
2) Why is the mapping quality so high when there are so many mismatches?
3) Is the reported mapping quality at the individual read level or at the pair level?
4) If I use the -s option to disable Smith-Waterman for the unmapped mate will it still use the results of the BW algorithm to get the reads in the pair that map?

[nilshomer]

1) Why does bwa try and rescue reads that don't map well with bwa aln?
Since repetitive hits (or other artifacts) can cause "bwa aln" to miss a valid alignment, so using one end as anchor for the other can improve both sensitivity and specificity.
2) Why is the mapping quality so high when there are so many mismatches?
I am not sure. Could you give an example with the before/after and what you expect?
3) Is the reported mapping quality at the individual read level or at the pair level?
Both, it is is the original mapping quality, adjusted based on how well the pair maps.
4) If I use the -s option to disable Smith-Waterman for the unmapped mate will it still use the results of the BW algorithm to get the reads in the pair
Yes, but try it yourself on a small # of reads to confirm.