This is probably the best known of all bioinformatics applications, and consequently has various different aspects to it.
General blast issues
The nr and nt databases are up to date and reside at /shelf/public/blastntnr/preFormattedNCBI this directory is available on all the nodes.
Because they are the preformatted versions, all you have to do is specify "nr" or "nt" as the database and remember to put this into your .ncbirc file in your home. i.e. cat ~/.ncbirc
[BLAST] BLASTDB=/shelf/public/blastntnr/preFormattedNCBI BLASTMAT=/shelf/public/blastntnr/ncbidatadir
This version which stopped development in 2010, used MPI to parallelise the blast process, by splitting up the database itself and running the query on the parts in parallel, roughly speaking. During 2015, Jens Breitbart made some modifications to the code, and called it mpifast (despite the fact that the underlying executable is still called mpiblast).
Therefore the database need to be fragmented and also - as is usual in blast - formatted.
The number of fragments is two less than the number of processes to be used. So, for 64 processes, the database will need to be fragmented into 62 parts. A script for doing this is as follows:
#!/bin/bash #$ -cwd #$ -j y #$ -S /bin/bash #$ -V #$ -q all.q module load mpifast export BLASTDB=/shelf/public/blastntnr/mpiblast46frags export BLASTMAT=/home/DatabasesBLAST/data export MPIBLAST_SHARED=/shelf/public/blastntnr/mpiblast46frags export MPIBLAST_LOCAL=/shelf/public/blastntnr/mpiblast46frags gunzip -c /shelf/public/blastntnr/nr.gz >./nr mpiformatdb -i nr -N 78 -t -p T rm -f nr
As you can see, this uses up a good deal of temporary hard disk space. There is an alternative way by using "zcat" and a pipe, but this also names the database fragments as "stdin" which is quite inconvenient.