[Dock-fans] discrepancy between rigid and flex
John J. Irwin
jji at cgl.ucsf.edu
Mon May 4 17:19:30 PDT 2009
Francesco Pietra wrote:
> I wonder whether a large discrepancy between the docking region
> between rigid and flex dock is a sign of some artifact in setting up
> the procedures. This implies that i have carefully tried to discover
> artifacts at no avail.
The only reason I see for using rigid docking is to investigate whether
there is a sampling problem with the cognate ligand or possibly a
problem with scoring. I see no reason to ever perform rigid docking
with any ligand other than to redock the crystallographic ligand. Just
want that to be clear!
> For example, with same input files, changing
> from allHIP state to allHIS state of the protein (and removing the
> extra proton of HIS from the pdb file) leads from a normal situation
> (i.e., rigid and flex differing mainly in the ligand conformation) to
> an absurd situation of rigid and flex poses being located nearly as
> far apart as the range of the protein could allow, while the
> conformational difference is the same as from the allHIP state of the
Changing HIS to HIP/HID/HIE (protonating or explicitly deprotonating the
histidines) should be guided on a case by case basis due to the local
environment if possible.
> I understand that without showing the files one can't expect much from
> such a question, however, i am asking about general experience
> suggesting what may be grossly wrong under the described observations.
If you fail to recapitulate the crystallographic ligand pose, the first
thing to ask yourself is: is this a sampling problem or a scoring
problem? You investigate this using a reductionist approach. e.g.
score the crystallographic ligand as observed in the crystal structure
compared to the best docked pose. Which scores better? Then, retain
the top N poses. Do any of them resemble the crystallographic ligand?
Break down the score on an atomic basis to get at why the pose you get
is unexpected. If necessary, break the ligand up into smaller moieties
and see how they dock. Conformational sampling is often the culprit in
molecules with large numbers of rotatable bonds, and fragmenting the
molecule can help to get at this problem.
> If one asks what is likely to be the HIS/HIE/HIP state of the protein
> i can only answer that the complexity is such that pKa-solving
> programs got confused, proving of no help.
It is wise to be cautious about automatic pKa prediction programs since
pKa is very subtle. Your best bet is to a) look at each site and figure
out HIE/HID/HIP by hand, if you can. b) if ambiguous and outside the
binding site, leave as HIS (or you can create subtle overall charge bias
if you think that will help) c) if ambiguous and in the binding site,
you may need to sample both HI[D;E] and HIP explicitly in separate
docking runs, although you may be able to see that one is clearly
favored based on available actives.
Take nothing for granted and set up hypotheses that can fail in
interesting and informative ways.
Hope this helps.
UCSF DOCK Team
> francesco pietra
> Dock-fans mailing list
> Dock-fans at docking.org
More information about the Dock-fans