[Dock-fans] grid parallelization?
chiendarret at yahoo.com
Tue Nov 13 23:47:50 PST 2007
At any rate, the grid was built in a few hours (maybe 4-5hours). With 10A in
the showbox, now the spheres and grid seemed to encompass the whole pore
protein. Rigid dock (mpi working with only two nodes instead of the four
requested) completed in about two hours for a 150-atoms stretched ligand. This
is now elongated within the pore. far end from the membrane. In fact, 1/3 of
the ligand is out of the pore. Whether this far-end position is the result of
selecting the center of the sphere on the oxygen of the HOH residue (which is
on the fourth position of the selectivity filter and from visualizing programs
appears along the symmetry axis of the pore) can't say. I would like to rebuild
the grid centering sphere_select more toward the membrane opening of the pore,
though I have no symmetrical position to select. I could only select an atom on
the internal helix, which would destroy symmetry. Don't know if this would be
>From vague descriptions, it seems that GOLD program can select a point along
the symmetry axis of the pore. Not read the original description how that is
realized, if it is really realized at all.
--- Scott Brozell <sbrozell at scripps.edu> wrote:
> On Tue, 13 Nov 2007, Francesco Pietra wrote:
> > I wonder whether the calculation of grid is suitable to parallelization in
> > future version of DOCK.
> > Using Magis' sphgen_cpp I can treat a whole pore protein model. I
> > today that I has spheres for only a portion of the protein. Therefore,
> > Magis' sphere_select, I have now selected a radius around a central atom
> > the pore (the oxygen of a "natural" HOH residue). Because I am exploring
> > without previous knowledge where the ligand prefers to dock, I am creating
> > grid for a box measuring 66 66 66 A for a total of more than 10 millions
> > points. In 30 min 10% of protein atoms processed. All processors, except
> > (100%) are dormant.
> > That of having a point of symmetry is another problem. Luckily I had that
> > oxygen of WAT residue in the receptor noH.pdb file.
> Thanks for the suggestion; we'll add it to the list.
> However, the flip side is that really big grids have to be stored and
> then read by program dock. The divide and conquer approach suggested in
> the tutorials is tractable and amenable to very coarse grained
> parallelization for both grid and dock, ie parallel runs of grid and dock.
> An alternative in the era of the approaching petascale computing
> is gridless DOCKing.
Be a better pen pal.
Text or chat with friends inside Yahoo! Mail. See how. http://overview.mail.yahoo.com/
More information about the Dock-fans