How does bowtie2 calculate read mapping qualities?

[mmanhart]

I'm wondering exactly how bowtie2 calculates the read mapping qualities shown in the output SAM file. The documentation makes it very clear how the internal mapping score is calculated, and how you can customize it (e.g., setting various penalties for mismatches and indels), and it says that this is used to calculate the actual mapping quality, but I can't find anywhere that shows exactly how the mapping quality is ultimately calculated.

Does it use the same heuristic formula as in MAQ? Here's the paper I know about:

Li H, Ruan J, Durbin R. (2008) "Mapping short DNA sequencing reads and calling variants using mapping quality scores." Genome Res. 18:1851--1858.

Many thanks for your help,
Michael

[mmanhart]

Can anyone here help me out? I would really appreciate it!

Thanks,
Michael

[sdriscoll]

the MAPQ isn't super informative. only got to do with the uniqueness of the alignment where higher scores means more unique. so it's probably a -10*log10(p) where p is some probability of the alignment being unique. I don't know what their probability formula is, however. if this is what they do then the formula isn't a straightforward one like they use in Tophat and STAR.

[mmanhart]

I know what you're saying --- that's all I can find documented in the Bowtie2 manual and publisher paper. However, I would indeed like to see the explicit formula used to calculate this. I find it really striking that it is not documented clearly --- many downstream filters work on mapping quality, so it should be very improtant to know exactly what goes into it.

[sdriscoll]

You're right..they should at least put it into words so one can know what they are doing.