Thanks to all who responded to my question regarding the energetics of a 
known interface applied to orthogolous dimers.

Steven Darnell asked me for some clarifications.  I have the structure 
of a homodimer, defined the dimerization interface and substituted the 
residues at said interface with those of each of four human orthologs of 
the original yeast protein.  What I call monomeric hybrids are thus 
yeast proteins with different humanized dimerization interface.  I 
recombine these monomeric hybrids to give four humanized homodimers and 
six humanized heterodimers.  It is the likelihood that these dimers form 
that I'm interested in.

Following Diana Tomchick's suggestion, I had PISA analyze the interface 
of the original dimer and learned  that it might be worth to consider 
one other region besides the main helix.

For modeling of the monomeric hybrid, MODELLER is suggested because it 
does simulated annealing by default.  To get the energetics, Rosetta 
with the "-interface" or "-ddg_only" flags might be a good tool.  I have 
to read up on the details.  An alternative is molecular dynamics in Gromacs.

Thanks for your help.


Andreas



Steven Darnell wrote:
> Andreas,
> 
> Here's my $0.02.  Would you mind clarifying a few things for me?
> 
>> I am working on (the theoretical side of) a protein complex whose 
>> structure has been solved.  The protein homo-dimerizes, mediated 
>> primarily by two long helices.
> 
> So you have a structure of a homodimer...
> 
>> Using sequencing alignment and the WHAT IF server, I built monomeric 
>> hybrid models containing the bulk of the known structure and the 
>> dimerization helices of homologous proteins.  Naturally, I want to 
>> know how likely they are to form dimers.
> 
> Could you explain what you mean by "monomeric hybrid"?  I'm guessing you
> want to thread two copies of monomer B onto the backbones of homodimer A.
> 
>> To look at the energetics, I've run the phenix geometry regularization 
>> algorithm to minimize clashes and side chain energies. 
> 
> I've never used phenix and I don't know what sort of search function it
> uses.  If it's a deterministic algorithm, like dead end elimination,
> you'll get the global minimum energy conformation with one run (if it
> converges, that is).  If it's a stochastic algorithm, like Monte Carlo,
> you'll never know if you're at the global minimum.  Your best bet is to
> run multiple independent minimizations, say 50-100 for starters, and
> pick the conformation with the lowest energy score.  I'm betting its the
> latter.
> 
>> The backbone conformation only changes minimally.  Next I calculated 
>> in Rosetta the energetic scores of the models before and after 
>> regularization and compared with that of the native structure.  This 
>> gave me some numbers that are not inconsistent with experiments.
> 
> The following assumes I correctly stated your design problem.  Rosetta
> does not account for conformational entropy, so the closer the backbones
> are between the homodimer A and modeled homodimer B to one another, the
> better.  You might want to consider fixing the backbones during
> minimization.
> 
> Also, I don't understand the purpose of calculating the energy of the
> non-optimized structure.  I would be more interested in the change in
> binding energy between the bound and unbound state of the minimized
> structure.  Rosetta can calculate that in "-interface" mode.  There's a
> flag to keep Rosetta from performing any design calculations; I think
> its "-ddg_only" or something like that.  Note that this calculation
> assumes the monomers behave like rigid bodies.
> 
> Finally, I would minimize homodimer A the same way you minimize modeled
> homodimer B as a control, then use Rosetta to calculate its change in
> binding energy.  Side chain flips of His, Asn, and Gln will make a big
> difference.  This will give you a number to compare to your modeled dimer.
> 
>> Before I sit down and write this up, I wanted to ask the community if 
>> what I've done makes sense and if there are alternative methods for 
>> minimizing and calculating interface energies.  I don't necessarily 
>> need docking algorithms as the interface is known.  I just want to get 
>> an energetic description.
> 
> If it were me, I would create the homology model using MODELLER (it uses
> simulated annealing by default), minimize/relax the structure using
> Rosetta, then calculate the change in binding energy with Rosetta.
> Remember to repeat stochastic processes.  The 50-100 time guideline was
> given to me by Deanne Sammond, as in:
> 
> Sammond DW, Eletr ZM, Purbeck C, Kimple RJ, Siderovski DP, Kuhlman B.
> Structure-based protocol for identifying mutations that enhance 
> protein-protein
> binding affinities.
> J Mol Biol. 2007 Aug 31;371(5):1392-404. Epub 2007 Jun 8.
> PMID: 17603074 [PubMed - indexed for MEDLINE]
> 
>> Thank you.
> 
> No worries.  Does any one else have any suggestions or corrections?
> I've only had 7 hours of sleep since Saturday.
> 
> ~Steve
> 
> -- 
> Steven Darnell
> Univeristy of Wisconsin-Madison
> Madison, WI USA
> 
> [log in to unmask]
> 

-- 
         Andreas Förster, Research Associate
         Paul Freemont & Xiaodong Zhang Labs
Department of Biochemistry, Imperial College London