sbrozell at scripps.edu
Fri Sep 14 08:38:42 PDT 2007
On Thu, 13 Sep 2007, Francesco Pietra wrote:
> Hi Scott:
> Thanks for your illuminating comments. Based on them, may I ask for details?
> Please see below.
> --- Scott Brozell <sbrozell at scripps.edu> wrote:
> > On Wed, 12 Sep 2007, Francesco Pietra wrote:
> > > As a potential newcomer to DOCK6 (for different purposes than
> > pharma-scoring),
> > > two questions related to problems I am faced with:
> > >
> > > (1) Is it any way to use opemmpi for parallelization (that I use for
> > Amber9,
> > > both compiled with Intel) in place of MPICH2? Having limited experience
> > with
> > > such tricks, I foresee major problems for me. In particular if the code is
> > > Fortran, where Intel defeats GNU compilers.
> > Yes. It might be as simple as pointing MPICH_HOME at the Open MPI
> > installation. My advice is to use gnu compilers to install
> > both Open MPI and DOCK. See the TROUBLESHOOTING section of
> > DOCK config file install/gnu.parallel for other comments.
> > This stale thread might be helpful:
> > http://blur.compbio.ucsf.edu/pipermail/dock-fans/2006-November/000776.html
> As Amber9 runs faster on Intel Fortran than GNU Fortran compilers (at least
> using standard libraries, as in my case), I wish to leave that as it is. It
> works very satisfactorily. Therefore, I have to check if my computer can have
> two different compilations of openmpi, the existing one with Intel and the new
> one with GNU compilers (toward which MPICH2 should be directed).
Note that DOCK installs its own Amber routines (which are actually NAB).
Since Intel and GNU are link compatible, GNU compiled DOCK
linked with Intel compiled Open MPI should work. But one can have
multiple Open MPI installations.
> > > (2) Is it productive to use amber scoring? GB with Amber9 for large systems
> > is
> > > unproductive with respect to explicit solvent-PME. Not to say about all
> > > pitfalls by implicit solvents.
> > Amber score has been used productively for rescoring;
> > the computational cost increases as the size of the movable
> > region increases. The cost of movable=nothing is similar
> > to that of the other non-grid based scores.
> > For a large receptor movable=everything will be slow and
> > thus may not be productive but still can be useful.
> > The movable=distance and movable=nab_atom_expression
> > of Amber score enable detailed control of receptor flexibility.
> > Many interpretations can be made of the Amber9 benchmarks:
> > http://amber.scripps.edu/amber9.bench1.html
> > In general, I won't disagree that explicit solvent-PME is better than
> > GB for long MD on very large systems, but if you want long large MD
> > then use Amber not DOCK. If you want to dock after you have done
> > long large MD then use DOCK especially if you want to apply a
> > uniform protocol with some small degree of specific receptor flexibility.
> That of running DOCK after amber's md is an interesting suggestion and I
> foresee usefulness also in my cases. The plan I am considering in studies of
> the interaction of certain natural products (ligands) with proteins (receptors)
> is the following:
> (1) Running DOCK in the simplest and fasted way to search if there is any
> docking at all. That is intended to build a complex ligand-receptor for a
> specific ligand with potential receptors.
> (2) Running Amber extensively for the ligand-receptor complex identified by
> (3) Running DOCK again on the results from (2).
> All that because I find difficult to start with Amber for my potential ligands
> and receptors that are only selected on chemical intuition. I just don't know
> how to build a reasonable system for Amber (although I am able to provide Amber
> with compatible FF and coordinates for my potential ligands). A question posed
> to Amber to this regard remained unanswered, either because it is a trivial
> problem for most or because of the opposite.
> All these are computationally very demanding tasks, so that I would be much
> obliged for comments on my plans. As I said in my previous email - and is
> implied here - I am not looking for potential drugs, although suggestions to
> this regard might arise as by-product of my studies.
This is a reasonable general plan.
For 1 use grid based scoring; follow the first four tutorials at
2 is consistent with my belief that thorough receptor preparation
and serious MD are necessary but may not be sufficient for binding
energies; see my continual comments on these issues:
On 3, redocking a family of ligands to a well prepared receptor and then
rescoring with Amber score is an intended application of Amber score:
Comments from Amber score users and citations to publications
More information about the Dock-fans