Thread: BFAST published
View Single Post
Old 11-11-2009, 12:26 PM   #5
Nils Homer
nilshomer's Avatar
Location: Boston, MA, USA

Join Date: Nov 2008
Posts: 1,285

Originally Posted by lh3 View Post
I did not read into details, but something is a little bit odd in Figure 3. How can bowtie and bwa produce wrong alignments even if the read can be perfectly mapped? For perfectly aligned reads, both of them should always give correct results. Also note that 6% error is really a lot. If an aligner was that bad, it would not survive these days. Nonetheless, I agree that the performance of bwa degrades with higher error/mutation rate.
What panel are you looking at (sensitivity or false mapping)? Obviously the sensitivity cannot be 100% due to the repetitiveness of the human genome. For false-mapping, with no errors I don't see a >0% false mapping rate in Figure 3.

In SNP calling given short reads, alignment global to reads has some advantage. Suppose the 3rd base in a read is a true mutation. Local alignment will clip the read and as a result you see more reads having the reference allele than having the non-reference allele, which leads to strong reference bias. To alleviate this bias, we need to postprocess local alignments.
We don't clip the reads. The "local alignment" we use is a global alignment to a window the candidate location. In other words, we enforce that the entire read must align to some subset of bases in the reference.
nilshomer is offline   Reply With Quote