I am trying to map reads from whole genomes, and considering whether to employ older BWA-aln or newer BWA-mem aligners. BWA-mem will supposedly map more reads than BWA-aln, sometimes this difference can be huge (80% vs 18%)
https://www.biostars.org/p/94742/
and consequently coupled to GATK BWA-mem gives more raw SNPs and less SNPs that have passed GATK VSQR filtering (probably low quality due to forced mapping).
https://peerj.com/articles/421/#table-3
Difference is not so big but basically BWA-aln gives ~30k more SNPs that passed filtering. Should I use older BWA-aln?
BWA-mem is "very good" at aligning almost everything. But this is not always "good". For instance, with samples that have low human endogenous content, you can see that the proportion of mapped reads is much larger with MEM than with ALN. In such a situation BWA-aln is a better choice because BWA-mem will be aligning reads which are not of human origin.
BWA-aln is more sensitive to the presence of adaptors and barcodes and this may also affect the proportion of mapped reads with ALN and MEM. The choice between ALN and MEM depends on many factors, although in general MEM should work better.
In your case, I would recommend to get the set of reads mapped only with MEM or ALN and look at which properties they have and how well they aligned. This may help spotting out what's going on with your particular sample.
I would run them both on your data if it is a concern. BWA runs very quickly in my experience so it shouldn't be an issue to run multiple aligners.
The advantage of bwa-mem is that it supports not just an end-to-end alignment, but local (portions of the read/ contig) or chimeric alignments. So it should run slower but produce more successful mappings. But possibly a greater percentage of ambiguous mappings(??)... If you have short read lengths, than I wouldn't think there would be much advantage to bwa-mem, since it is unlikely that you require anything other than an end-to-end match (i.e. no large gaps) and thus could take advantage of the faster bwa-aln. I've attached a link to the paper describing the bwa-mem algorithm.
You might also try some other burrows-wheeler aligners, like BBmap or Bowtie. I usually use BBmap- it is worth looking in to even if you don't end up using it, for all of the useful tools and scripts that come with it (http://sourceforge.net/projects/bbmap/).
http://arxiv.org/pdf/1303.3997v2.pdf
Thanks for you answer. Yes that is my main concern increased percentage of ambiguously mapped read or just missmapped reads with BWA-mem, that will thus give more low quality SNP calls. This is something what we see from the table I attached, (BWA-aln gives 30k more SNPs that passed filtering than BWA-mem).
I know BWA-mem is better for longer reads (>70bp), meaning better % of the mapped reads but my concern is whether this is actually needed for SNP calls, as it could increase number of false calls.
BWA-mem is "very good" at aligning almost everything. But this is not always "good". For instance, with samples that have low human endogenous content, you can see that the proportion of mapped reads is much larger with MEM than with ALN. In such a situation BWA-aln is a better choice because BWA-mem will be aligning reads which are not of human origin.
BWA-aln is more sensitive to the presence of adaptors and barcodes and this may also affect the proportion of mapped reads with ALN and MEM. The choice between ALN and MEM depends on many factors, although in general MEM should work better.
In your case, I would recommend to get the set of reads mapped only with MEM or ALN and look at which properties they have and how well they aligned. This may help spotting out what's going on with your particular sample.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics
- Bioinformatic Tools