Incorporate mapping scores into mutation calls

[GoogleCodeExporter]

Will probably need to use raw base qualities to estimate these mapping score 
differences. (Using re-calibrated errors would be not quite correct, and 
require too much re-processing.)

As a first step, can incorporate the number of mismatches (currently recorded) 
as a uniform error to mapping (perhaps with different value for indels). This 
would catch the worst offenders.

Original issue reported on code.google.com by jeffrey....@gmail.com on 12 Jun 2012 at 3:25

[GoogleCodeExporter]

With change to Bowtie2, can use true mapping qualities. Only complication is 
that we need to estimate these for junction aligned read fragments by comparing 
the alignment to the junction to the best possible alignment score to the 
reference genome.

Need an example data set where this sort of problem occurs. Could generate one 
using manB-cpsG region or rrlA gene. 

Original comment by jeffrey....@gmail.com on 29 Jul 2012 at 4:49

[GoogleCodeExporter]

Done. This does get rid of some noise in this regard among the marginal RA 
evidence with scores as high as 7.5 in ZDB30.

Currently, junction-mapping reads are arbitrarily assigned perfect mapping 
scores when they are resolved to the reference genome. This is not quite right, 
but it's not clear how to calculate consistent mapping scores short of 
re-aligning them to just the accepted junctions and the reference with Bowtie2 
again. (This would be fast, so maybe it is the best option.)

Mapping scores could also be used to assign effective coverage, instead of 
simply the number of reads mapping somewhere.

Original comment by jeffrey....@gmail.com on 1 Aug 2012 at 9:08

[GoogleCodeExporter]

Still need to add this to polymorphism calculations however.

Original comment by jeffrey....@gmail.com on 1 Aug 2012 at 9:09