BWA -n flag using FLOAT

[mscholz]

All,

I am having difficulty deciphering my results.

Briefly, when I use -n (FLOAT) during bwa aln, my expectation is that this is analogous to the number of mismatches/seed using a sliding scale, rather than a hard integer number of mismatches.

I am starting to suspect that this is incorrect.

When I run a range of -n (FLOAT) from 0.02 - 0.1, I get more reads mapping to my reference using .02 than .1.

Can someone explain exactly what -n FLOAT is doing?

Thank you.

[nilshomer]

The FLOAT option states "missing prob under 0.02 err rate". Therefore, you are giving the maximum acceptable probability of missing the correct alignment, under the assumption of a 0.02% error rate. A larger value (0.1) will make it more likely to return an incorrect mapping, but be a lot faster. A smaller value (0.02) will more sure that a mapping is not missed, but at the cost of speed.

[ichorny]

So is there a way in BWA to allow reads to align up to a fixed error rate say 4%? Seems like the -n option limits you to 2% max. Is that correct?