TopHat (Tuxedo Suite) - AWS vs Local

[drsnafu1831]

Hello All
Sorry this one's a bit long
Have come here after some testing !!

I am trying to run Tuxedo suite for RNASeq Analysis on AWS

Post alignment using Tophat, I have recorded significant difference in size of "accepted_hits.bam" between runs on AWS (Amazon Web Services) and Local System

Details
Paired end FastQs
Size - after Adapter and quality thresholding - 7 gigs (3.5 X 2)
Human Sample

AWS Instance
15 cores - of - m4.16xlarge (64 cores/256 RAM)
Time ~45 min
Out put accepted_hits.bam ~100mb

Local Server
15 cores - of - 36 cores/256 RAM
Time ~52 min
Out put accepted_hits.bam ~1.2GB

Questions
Why such significant difference? reasons?

One thread I found, which was somewhat related
https://www.biostars.org/p/102874/

As suggested in the above thread, I have tried running tophat, with same parameters and input files,
on 1, 5, 10, 15, 20, 30 and 64 cores on local servers
on 30 and 64 cores on AWS

Interestingly, the size of "accepted_hits.bam" remained same (1.2 gb) till I reached 30 cores on local server (with specs mentioned above), and reduced (~120 mb) on 64 cores.
On AWS, as said above, 30 and (later) 64 are giving out ~100 mb of accepted_hits.bam

any input, suggestions and comments are welcome
thank you for your time !!

[Brian Bushnell]

Sounds like a bug to me... maybe you should try a different version of Tophat. The number of cores should not affect the size of the output bam more than a tiny amount.

Alternatively, you could try BBMap; it's faster than Tophat and produces correct output for any number of cores