MPI-HMMER is primarily a
parallel implementation of the two search functions of HMMER,
hmmsearch and hmmpfam. The other utilities within the HMMER
suite are still available, but are not currently implemented
within MPI. We assume that the user has a working MPI
implementation and has either compiled MPI-HMMER from source
or has installed one of the prebuilt RPMs. Here we detail
options that are specific to MPI-HMMER. For non-MPI options
(such as the HMMER options) we refer the reader to the HMMER
The hmmsearch utility
functions similarly to the PVM
implementation. As an example, consider a search using 16
16 -machinefile <machines> ./hmmsearch
--mpi /path/to/HMM/file /path/to/sequence/database
Most of the output is
formatted the same way as HMMER's.
However, we make some modifications to the header.
Because hmmpfam is more IO
intensive than hmmsearch, we retainthe requirement that the
database be made available to all nodes participating in the
computation. This is typically done with a network file
system, but can be local to all nodes if needed.
First, the HMM database must be indexed:
Just like in the PVM
implementation, this will create an index
of the HMM file that you intend to search against. This step
only required the first time you use a particular database.
Subsequent searches can reuse the resulting .ssi file.
To run hmmpfam, perform the
mpirun -np 16 -machinefile <machines> ./hmmpfam
--mpi /path/to/HMM/database /path/to/search/sequence