Try to further accelerate Bowtie2 alignment

[Tonnny97]

Hi all,

I try to further accelerate Bowtie2 alignment by separating input file to several pieces and using Hadoop-like computing framework to execute.
Based on my understanding, Bowtie2 aligns each pair of read independently.
However, when I separate input files into two parts and merge two outputs of Bowtie2, the content of result file is somewhat not consistent with that of result file generated without separating input.

I would like to ask whether Bowtie2 has some dependent task inside?
Does separating files and then merging the output make sense in Bowtie2?
Thanks.

[dpryan]

I've noticed that happen on occasion and still haven't really found the cause. My presumption is that the random seeding will be slightly different if you split the files (if the random number generator is always given the same seed, then splitting up a file will result in different numbers being produced if you split reads over files, as opposed to the completely reproducible results if you run an alignment twice). Interestingly, apparently a similar thing happens in BWA if you use different numbers of threads. This was discussed over on Biostars and no one's come up with the base cause yet.