Benjamin Berman, University of Southern California
7 May 2010
In your comparison to MAQ, you state that "MAQ keeps non-uniquely mapped reads and assigns them to one of the best-matching positions". This is true, for any read that maps to multiple places in the genome, MAQ does output it. But it also outputs for every read a phred-scaled probability score for the read being correctly mapped. The vast majority of ambiguous or repetitive reads get a score of 0 since they can't be mapped uniquely. This is not unique to Maq - many aligners output ambiguous reads because they can be used for some downstream applications (like RNA-seq). Most MAQ users filter out non-unique mappings (with scores less than 20-30, and indeed most of the repetitive reads have a score of 0). We routinely use MAQ for bisulfite and non-bisulfite sequencing, and would never use the results without first filtering based on mapping score - as you point out, including these ambiguously mapped reads would add a huge amount of noise to the methylation/SNP calls.
If you had filtered based on mapping quality, you would have found the MAQ numbers to be much more comparable with those of RMAP and BS SEEKER. By including the ambiguous or poorly mapped reads, it's an apples to oranges comparison that is kind of meaningless.
MAQ and uniquely mapped reads.
7 May 2010
In your comparison to MAQ, you state that "MAQ keeps non-uniquely
mapped reads and assigns them to one of the best-matching positions". This is true, for any read that maps to multiple places in the genome, MAQ does output it. But it also outputs for every read a phred-scaled probability score for the read being correctly mapped. The vast majority of ambiguous or repetitive reads get a score of 0 since they can't be mapped uniquely. This is not unique to Maq - many aligners output ambiguous reads because they can be used for some downstream applications (like RNA-seq). Most MAQ users filter out non-unique mappings (with scores less than 20-30, and indeed most of the repetitive reads have a score of 0). We routinely use MAQ for bisulfite and non-bisulfite sequencing, and would never use the results without first filtering based on mapping score - as you point out, including these ambiguously mapped reads would add a huge amount of noise to the methylation/SNP calls.
If you had filtered based on mapping quality, you would have found the MAQ numbers to be much more comparable with those of RMAP and BS SEEKER. By including the ambiguous or poorly mapped reads, it's an apples to oranges comparison that is kind of meaningless.
Competing interests
None declared