How accurate is BWA MEM mapping?

Question

[hah0024]

I am trying to compare different databases and the amount of reads that map from my sample, being a collection of short reads from the environment. After looking in depth in the manual, they say the BWA MEM is "accurate", but how accurate is it actually? How close does a sequence have to be to be considered a match to the reference? Additionally, can anyone explain the difference between the -T option, and the -A option? And what the 'match score' is? Is this related to % identity?

Thanks for all the help in advance!

Answer 1

[mforde84]

The only way you could assess accuracy is if you already knew the consensus sequence. Alignment sensitivity and specificity is an issue addressed indepth for clinical variant calling. I might suggest looking into Genome In A Bottle (GIAB). If I understand correctly, bwa-mem is the defacto standard currently for reads lengths >= 50bp.