Why does MMSeq advises against the use of Bowtie2 ?

Question

[Thibault D.]

Hi,

As I am about to use MMSeq and MMDiff for differential expression analysis in RNA-Seq, I have been looking at the github repository. I have been reading all MMSeq advices and understand their choices.

However, I do not understand: why should Bowtie1 be used and not Bowtie2 ? In fact, when I use Bowtie1, I have 49% of mapped reads (35% with default parameters) ; by contrast, I have a 97% of mapped reads with Bowtie2 (corresponding parameters have the same value) and 95% with default parameters. Please, note that each alignment have been performed against transcripts (fasta format) and that reads are under fasta format (no fastq quality scores).

Up to now, I have not tried to run MMSeq from Bowtie2, as I do not understand why it should not be done.

Would you mind helping me ? Thanks in advance !

Answer 1

[Devon Ryan]

The reason they suggest people use bowtie1 vs. bowtie2 is that you can have bowtie1 output an unlimited number of equally good hits, while this isn't possible with bowtie2. Having said that, since you're so obviously getting better alignments with bowtie2 I would recommend that you ignore what they suggest and just use bowtie2. Note that you'll need to filter the results to ensure that all of the alignments passed to MMSeq have the same (or close to the same) alignment score. You could do this with a little python script.