[Dock-fans] Dock is too slow even in parallel using MPI
jason.phan at Vanderbilt.Edu
Tue Apr 17 10:13:59 PDT 2012
Thanks for the suggestion and sample input file. We're in the process of upgrading to dock6.mpi. I cut down everything to a bare minimum e.g. max_orientations 25, min_anchor_size 40, num_anchor_orients_for_growth 50, num_confs_for_next_growth 50, simplex_anchor/growth_max_iterations 100. I have a very well-defined pocket for a big ring so I don't need too many initial orientations. So far the energy score comes down a little bit and more rejections due to lack of orientations but the speed is up a lot, ~10x. I'll view the poses when it's done to see if it makes sense.
On Apr 17, 2012, at 8:35 AM, Trent E. Balius wrote:
My first suggestion is to upgrade to DOCK v6.5 the latest release which has the latest features and code improvements.
1-3 min per ligand seams about right for on-the-fly growth in DOCK5 and DOCK6.
Both DOCK 5 and 6 are slower than DOCK 4, but have a higher success rate in pose reproduction experiments.
Note that the MPI implementation of dock is designed for virtual screening applications. DOCK will distribute molecules to different nodes. DOCK does not parallelize the growth of individual molecules.
See the following:
The other parameters that impact the speed (but also sampling) are:
min_anchor_size 40 only one anchor is used. min_anchor_size 5 all rings are used.
pruning_max_orients is the maximum number of anchor orientations carried forward in the anchor & grow search (previously num_anchor_orients_for_growth)
pruning_clustering_cutoff is the number of orients retained at each stage of growth referred to as N_c in the equation in Pruning the Conformation Search Tree.
The following is an example virtual screening script used by the Rizzo Group to good effect on a BlueGene.
I hope that this helps,
Trent E. Balius
Graduate Student, Rizzo Group
Department of Applied Mathematics and Statistics,
Stony Brook University.
Office: Math Tower 3-129, Phone: (631) 632-8519
From: Phan, Jason <jason.phan at Vanderbilt.Edu<mailto:jason.phan at Vanderbilt.Edu>>
To: dock-fans <dock-fans at docking.org<mailto:dock-fans at docking.org>>
Sent: Mon, Apr 16, 2012 8:41 pm
Subject: [Dock-fans] Dock is too slow even in parallel using MPI
What are the parameters that can be manipulated to improve the virtual docking
performance without compromising the quality of the dock results? My grid box is
pretty small, covering ~11 spheres and pretty much default values were used for
all the parameters. Using parallel processing with dock5mpi over 4 nodes, 8
processors each, the speed is terribly slow, e.g. 1-3 min per ligand. I tried
lowering the max_orientations, simplex iterations, 1 conformer written, no
clustering etc... and it improved some. I'm wondering if there are better ways
to speed things up a lot more.
Dock-fans mailing list
Dock-fans at docking.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Dock-fans