[Dock-fans] Fw: Re: Fw: Re: Fw: Re: Fwd: Re: Ligand Docking with DOCK6
sbrozell at scripps.edu
Tue Apr 22 10:12:04 PDT 2008
On Tue, 22 Apr 2008, Francesco Pietra wrote:
> As anticipated below, I have killed the serial procedure because (after 2526 min CPU time; nearly two full days wall time), the "top -i" and *.out file remained the same as after 100 min CPU time (situation reported below) and there was no way to check if the procedure was advancing or not.
> Now I have launched the same flex docking, continuous score in parallel (8 logical CPUs) with same *.in file and adding -v for verbosity.
> The *.out file started with 8 lines:
> Initializing MPI Routines ...
> and ends with
> Initializing Library File Routines..
> Initializing Orienting Routines ..
> Initializing Conformer Generator Routines...
> VERBOSE MOLECULE STATS
> Number of heavy atoms 3417
> Number of rotatable bonds 2120
> There are also *.out.# (# from 1 to 5) files.
> *.out.1 shows for verbosity:
> Number of heavy atoms 73
> Number of rotatable bonds 3.
> *.out.2 and *.out.3. *.out.4 and *.out. show:
> Number of heavy atoms 3417
> Number of rotatable bonds 2120.
> Dock is currently running on deb64
> 0 Molecule Processed.
> "top -i"
> Initially showed 8 lines, 7 for the CPUs. Rapidly the active CPUs are only two (rarely a third one appears with zeo %CPU and %MEM)
> PID (3736), USER (f), PR (25), NI (0), VIRT
> (7051m), RES (6.5g), SHR (4192), S (R), %CPU (100), %MEM
> (28.2), TIME+ (12), COMMAND (dock6.mpi).
> PID (3735), USER (f), PR (25), NI (0), VIRT
> (370m), RES (11m), SHR (4176), S (R), %CPU (100), %MEM
> (0), TIME+ (12), COMMAND (dock6.mpi).
> The two sets of data exchange frequently the PID number. The %MEM and RES are increasing, reaching at TIME+ 50 the values observed before for the serial run.
> I offer a comment of the sphere_select.sph size. It has little meaning alone. With a 106-atoms ligand, the largest selected_spheres.sph (10,536,750 points) is dealt with fine in grid scoring with box 12A. Under the same conditions, increasing the size of the ligand to 118 atoms, crash occurs as described. I must add that these are ring-fused compounds, so that the increased number of atoms pertains to rings which (as far as I understand) are treated as rigid bodies. So, this appears to me a curious trend. I have also to add that reducing the selected_spheres.sph to 7,675,200, in a 8A box, does not provide room enough to the ligand in grid scoring, although the spheres still cover most of the protein. Therefore, I do not understand your suggestion "If you are uneasy then try a much smaller calculation for verification and benchmarking." I am not uneasy, I am used at the problems of scientific research, though I can't imagine how to check the issues with a
> smaller calculations: the largest sphere_select.sph is OK for a slightly smaller ligand.
There are several issues here. First a clarification:
In all docking calculations (i.e., orient_ligand == yes)
receptor_site_file defines the active site(s).
grid_score_grid_prefix represents a special region of the receptor;
special in that contact or energy scores can be computed very quickly
for ligands that dock such that the whole ligand is in the region.
Grids are not used with continuous score;
grid_score_grid_prefix is not a Continuous Energy Score Parameter;
with continuous score cont_score_rec_filename is the receptor.
Second I am not sure exactly what you mean by points.
It is best to quote verbatim input and output items. Note that
relating a .sph file to a showbox margin and Total number of grid points
is only a crude way to gauge active sites and
receptor region sizes. Grids are not used with continuous score;
showbox margin and Total number of grid points are irrelevant.
Third if you are uneasy about blindly leaving a long calculation
that is producing no output running then ramp up to a long
calculation, perhaps starting with a scientifically useless calc.
The main problem was memory consumption; the total number of spheres
in the receptor_site_file is directly related to memory consumption:
Take a receptor_site_file and delete half the spheres from it;
if necessary patch up the number of spheres in the header: e.g.,
cluster 1 number of spheres in cluster 12
run dock measuring the memory usage;
if the calculation is too long then stop it; repeat this procedure.
Eventually, you will get to a quickly running dock calc (perhaps
terminating without actually docking the ligand).
After a couple of steps you will have a formula for memory use
as a funtion of the total number of spheres and a feeling for
Fourth maybe your ligand(s) do not DOCK to your receptor.
Adding rings may produce a ligand that cannot fit into a pocket
that was a previous active site. This is research ...
> It seems to me that there is little hope with present parallel procedure. It seems to go on as a serial run. Actually, the two-CPUs issue is what was regularly observed in the past, even for successful runs. I saw all CPU at work only for amber score. Though, I understand these are complex affairs. I can't rule out to be missing some obvious point, and hope my posting is not disturbing developers an audience. Is any hope that present parallel procedure may give some hint? Wait for your advice before killing it.
DOCK is only parallelized over ligands. With one ligand in
ligand_atom_file running in parallel is the same as running serially.
Start with serial DOCK on a small representative set of ligands.
Once you have useful serial
calculation then run a library of ligands in parallel.
> --- On Mon, 4/21/08, Francesco Pietra <chiendarret at yahoo.com> wrote:
> > From: Francesco Pietra <chiendarret at yahoo.com>
> > Subject: Re: [Dock-fans] Fw: Re: Fw: Re: Fwd: Re: Ligand Docking with DOCK6
> > To: "Scott Brozell" <sbrozell at scripps.edu>
> > Cc: "dock-fans" <dock-fans at docking.org>
> > Date: Monday, April 21, 2008, 2:10 PM
> > --- On Mon, 4/21/08, Scott Brozell
> > <sbrozell at scripps.edu> wrote:
> > > From: Scott Brozell <sbrozell at scripps.edu>
> > > Subject: Re: [Dock-fans] Fw: Re: Fw: Re: Fwd: Re:
> > Ligand Docking with DOCK6
> > > To: "Francesco Pietra"
> > <chiendarret at yahoo.com>
> > > Cc: "dock-fans"
> > <dock-fans at docking.org>
> > > Date: Monday, April 21, 2008, 1:30 PM
> > > Hi,
> > >
> > > On Mon, 21 Apr 2008, Francesco Pietra wrote:
> > >
> > > > Good news, I hope. Please see below. A quick
> > response
> > > would be much appreciated in order to decide whether
> > to
> > > continue or stop the calculation. Thanks, francesco
> > pietra
> > > >
> > > > --- On Thu, 4/17/08, Scott Brozell
> > > <sbrozell at scripps.edu> wrote:
> > > >
> > > > > First priority is to investigate selected_spheres.sph.
More information about the Dock-fans