View Single Post
Old 03-21-2016, 04:03 AM   #2
Markiyan
Senior Member
 
Location: Cambridge

Join Date: Sep 2010
Posts: 115
Default For the course level use blastx(p), for more refined - blastn

I may suggest trying blastx or blastp (if you've got good protein predictions) for the coarse level of closest neighbour finding.

For more refined use blastn on the DNA sequence (nt or similar db).
It is a good idea to create more specific database (esp nt one), and have them on dedicated SSD (esp. nvme or PCIe one).

On the performance side of things, give blastn 900bp or 1kb chunks, turn megablast on and upp the word size. "-n T -W 32".
For blastx - raise the expect to 1e-5 or 1e-10, and give it 450 bp chunks.

Also it is good idea to play with number of cpus per job (depending on number of cores per cpu, db/ram size).

PS: also there is a blast-like tool called diamond (used by AntiSMASH), which may be a bit faster in some cases, but the developers kept changing it's output format, so not as stable as blast, if you need to write your own parser.
Markiyan is offline   Reply With Quote