running TopHat with ~1.2 billion reads

[shurjo]

Any tips for the quickest way to get TopHat to complete alignment of a very large dataset (~1.2 billion 75bp Single end HiSeq reads)? I considered splitting it into 10 but then realized this would risk treating identical reads in the ten subsets as unique in the final file. I'm looking to keep anything that maps up to 10 locations, with up to three mismatches and three bp indels:

 -p 16 -g 10 --read-mismatches 3 --read-gap-length 3 --read-edit-dist 3

.

Thanks,

Shurjo

[ECO]

Split it, and dedup it later?

[pbluescript]

Do you have to use Tophat?
STAR will do split read alignment in a small fraction of the time that it takes Tophat.

http://gingeraslab.cshl.edu/STAR/

[severin]

GPU option

SOAP3-dp is very fast as long as you are ok with only up to 4 mismatches.

[shurjo]

All of these sound very logical. I'm going to step out of my TopHat-based comfort zone and try STAR and SOAP3-dp.

ECO, nice to see the forum doing so well. I've been away from this world for a bit but am coming back soon.

Thanks everyone,

Shurjo

[ECO]

Shurjo: nice to see you back here. Always wished I'd been able to follow our collaboration through. Good luck in your current work.