[Dock-fans] docking protocol

[Scott Brozell]

Hi,

In addition to this response
http://blur.compbio.ucsf.edu/pipermail/dock-fans/2008-May/001593.html
(which may not have appeared in your mailbox)

another developer writes:

We tested various cutoffs for the energy_cutoff_distance.  We basically
found that any portion of the receptor that you cutoff during this
portion of the calculation resulted in errors in the electrostatics,
with a much lower impact on the vdw energies.  This result makes sense
as electrostatics are a long-range effect and cutting off portions of
the protein result in problems, whereas vdw are a short-range effect and
thus are not as dependent on the far-ranges of the protein.   We
recommend 9999 to make sure the entire protein is encompassed
for the electrostatic calculation (as stated in the DOCK 5 paper) but it
is possible to do shorter distances when looking specifically at vdw
interactions.

For spheres, I typically try to set up around 75-120 spheres (40 is
actually a bit on the low side).  This range usually balances between
coverage of the active site (avoiding biasing toward a particular area)
and sampling too much of the surface (requiring too much out of the
matching algorithm).  This range was also the one used to set up
sampling parameters.  I am not sure how changing the number of spheres
will effect the sampling.  Also, I don't think we have really done a
systematic study of the number of spheres in any sampling regime.


On Mon, 12 May 2008, Scott Brozell wrote:

> On Thu, 8 May 2008, John J. Irwin wrote:
> 
> > You might want to contact the authors of that paper at U Maryland for 
> > advice.
> > 
> > Jerry Parks wrote:
> > >
> > > I'm interested in performing normalized vdW attractive energy scoring as
> > > described in J. Med. Chem. 2005, 48, 4586-4595. Can Dock 6.2 do this
> > > automatically, or should I just write a script to extract the vdW
> > > attractive energy from the output and normalize it myself?
> 
> There is no automatic support for this in DOCK.
> 
> > > Could you comment on the energy cutoff distance in grid.in? Is it
> > > reasonable to use a finite cutoff (rather than the 9999 used in the
> > > tutorial). Is there any general rule for selecting the cutoff besides
> > > systematically testing the convergence for various cutoff distances? How
> > > about max_orientations?
> 
> The default value of 10
> http://dock.compbio.ucsf.edu/DOCK_6/dock6_manual.htm#GridEnergyScoring
> is the one used in the publication, p507
> http://dock.compbio.ucsf.edu/DOCK_6/dock6_manual.htm#MengetalJCompChem1992
> But was probably chosen for computational time reasons.
> Now speed is unlikely to be an issue, and the whole receptor
> might as well be used.
> 
> > > Also, is there any sort of general guideline for choosing the "right"
> > > number of spheres? My binding site is defined by three residues that are
> > > known to be important from mutation studies. I selected spheres within 8
> > > Angstroms of those three residues, and then augmented those with
> > > additional spheres corresponding to xray waters in roughly the same
> > > region. There are a total of about 40 spheres, and the coverage of the
> > > binding pocket looks fairly complete to me. Does this sound sufficient?
> > > Could this be too many spheres?
> 
> Ensure that the active site(s) are covered.
> Probably sufficient.
> Probably not, too many spheres is usually relative to computational resources
> or to calculations focused on specific regions of the receptor.
> 
> It would be good to hear from others on these issues.
>