BWA -n option

[Rocketknight]

-n 0.10 doesn't mean that 10% of the bases in the read may have mismatches. -n with a float is the expected fraction of alignments that will be missed, assuming an error rate during sequencing of 2%.

In other words, -n 0.10 means you want the maximum number of mismatches set such that you will be able to map 90% of your reads successfully, assuming a 2% error rate.

For 60bp reads, -n 0.10 is much more stringent than -n 6, and so you are only getting very close matches when you use this option. The fact that BWA does not have to search for alignments with many mismatches when you set -n 0.10 also explains why the running time is much shorter.

[xuer]

interesting! who knows the answer?

[seq_lover]

I have been trying to use BWA to align my solid reads. In the manual it has been mentioned that you can use the -n option to limit the number of mismatches allowed between reference genome and the read. -n can be provided as an integer or float. In the latter case, the maximum edit distance will depend on the read length and equals to the float(n)* read length (I assume). My reads are 60 bp in length and somewhere I read that a minimum of 6-8 bp mismatches should be allowed for a solid read of length 60 bp.

I ran two jobs first with -n 6 as an integer and for the the second one I provided -n 0.10. I assume the it means that 10% the bases in the read may have mismatches.

1) Am I correct in assuming that -n 0.10 means that 10 % of bases in a read are allowed to mismatch(error or variants) with the reference genome?

Assuming my understanding about this parameter is correct and using the BWA on same data(10,000,000 reads) on same machine (16 processors), here is the interesting thing I noticed:

Case1: n = integer , running time = 7234 seconds, .sai (output file size) = 510 Mb
aln -c -n 6 -d 15 -l 25 -k 2 -t 15

Case2: n = float, running time = 1216 seconds, .sai (output file size) = 350 Mb
aln -c -n 0.10 -d 15 -l 25 -k 2 -t 15

2) Does anyone have any idea why there is so large difference in running time and why the second case produced more result? I repeated the comparison with many small datasets and found significant difference in running time and output.

Thanks