Is BWA-MEM as Good as BLASR for Aligning PacBio Reads? - Part 1
We downloaded raw reads and alignment file of human genome data released by Pacbio yesterday. It came from a postdoc research project by Mark Chaisson at the Eichler lab in University of Washington. Please note that his picture below is taken on one of those two rare days during Seattle summer, when the sun comes out :)
With his long experience as Pavel Pevzner’s student, author of BLASR alignment program as a PacBio employee and at the Eichler lab, Mark can provide far better perspective of the data than what we can do here. So, please excuse our lame attempt in using it to answer a few basic questions.
One question that often pops up in our mind is whether BWA-MEM is as good as BLASR for aligning PacBio reads on to a genome. BWA- MEM is the long read alignment program written by Heng Li. We covered it extensively in our earlier commentaries.
Mapping God Found Scientifically Dishonest by Anonymous Peer Reviewers
A Number of Informative Comments from Heng Li on BWA-MEM Aligner
We also found “An open peer review of bwa mem” by The Darling lab that is quite informative, and here is the informal response from Heng Li.
Over the last few days, we have been dismantling the code of BWA mem just like we did earlier for SOAPdenovo2, Minia and DSK (k-mer counting code from Rayan Chikhi). Today we will update the wiki page more to explain the files and functions within BWA. Overall, the mapping strategies of BWA-MEM and BLASR appear very similar. Both methods first find a number of short seeds (i.e. perfect match) between the query and the reference using BWT. Then they both use Smith Waterman to fill up the regions between the seeds. Therefore, it is natural to ask, whether their performances match for PacBio reads. The difference between the algorithms is that the seeds of BLASR are of constant size, whereas BWA-MEM tries to find seeds of maximal size. However, when you consider the fact that multiple constant sized seeds of BLASR can be combined into a super-seed, above differences should not affect the mapping quality. BWA-MEM uses combined forward-reverse index of the genome, which likely speeds up the seeding step.
Our question has three parts:
(i) How different are the programs in aligning long PacBio reads on to a reference genome?
(ii) Can the parameters of the BWA-MEM algorithm be changed from default to improve performance?
(iii) Can the code of the BWA-MEM be tweaked to improve performance?
In part 2 of this commentary, we will address those questions up to our highest level of incompetence.
