bowtie SAM mapq field

[rgregor]

I am trying to find out how bowtie mapq field is computed:

From the bowtie manual:

If an alignment is non-repetitive (according to -m, --strata and other options) set the MAPQ (mapping quality) field to this value. See the SAM Spec for details about the MAPQ field Default: 255.

If i map data with, for example, -m 10, i am allowing up to 10 multiple hits per read. How is the reported SAM mapq number connected with the number of multiple hits of an alignment? If a read has a single hit (from the manual), mapq = 255. If a read has 5 multiple hits, mapq=?

SAM documentation doesn't tell me anything more (only that it is phread-scaled):

MAPping Quality (phred-scaled posterior probability that the mapping position of this read is incorrect)

In the end i would like to do something like this:

samtools view -q 100 result.sam (filter out results with mapq <100)

tnx for any help,
Gregor

[carmeyeii]

Tophat/bowtie don’t report mapping quality values that are as meaningful as BWA, but there is some information in the mapping quality values tophat reports. Tophat yields 4 distinct values for its mapping quality values (you can do a “unique” count on the mapping quality field of any SAM file from tophat to verify this):



255 = unique mapping

3 = maps to 2 locations in the target

2 = maps to 3 locations

1 = maps to 4-9 locations

0 = maps to 10 or more locations.



Except for the 255 case, the simple rule that was encoded by the authors is the usual phred quality scale:



MapQ = -10 log10(P)



Where P = probability that this mapping is NOT the correct one. The authors ignore the number of mismatches in this calculation and simply assume that if it maps to 2 locations then P = 0.5, 3 locations implies P = 2/3, 4 locations => P = 3/4 etc.



As you can clearly see, then MapQ = -10 log10(0.5) = 3; -10 log10(2/3) = 1.76 (rounds to 2);

-10 log10(3/4) = 1.25 (rounds to 1), etc.

[carmeyeii]

I think it is safe to say that bowtie does the same.