Re: multiple FEP experiments in one NAMD-FEP run?

From: Peter Freddolino (
Date: Wed Mar 18 2009 - 15:42:27 CDT

Hi Sebastian,

if you think about what you're calculating with the FEPs... the free
energy changes that you calculate are properties of *the system*, not
just local properties of one monomer. Any attempts to separate out the
monomers from each other are doomed to failure because they interact
with each other (even if only through long-range electrostatics) and
with their surroundings. You *might* be able to get a first-order
approximation of the free energy changes due to the individual mutations
using the methods you've described, but you'd be making an untested set
of assumptions and risk running into serious troubles if there is any
significant influence of one site on another or the environment as a
whole. If you really can't afford to do all of your FEPs independently,
you should either use a better established approximation or cut down on
the number of mutations you expect to be able to consider.


Sebastian Stolzenberg wrote:
> Thanks so much, Peter,
> let's forget about the fixed environment then.
> What about doing several monomeric FEP experiments in one oligomeric
> run? I can argue that the monomers don't talk to one another, but
> stabilize each other structurally.
> My idea is to produce one complete oligomeric FEP run, such that
> \delta_G_oligomer=sum over all \delta_G_monomer values
> To dissect the several free energy differences from one another, I would
> post-analyze the FEP trajectories (collecting Hamiltonian values of each
> monomer-system to get PMFs). The advantage is that I can start the runs
> now, and meanwhile learn how to do the post-analysis.
> I already checked the separation of the multiple FEP mutation sites: in
> VMD, displaying atoms "within $R" Angstrom around each FEP site, the
> resulting "blobs" only start to overlap at $R~19A.
> In principle, is *this* reasonable? ... feasible with NAMD-FEP?
> Thank you again,
> Sebastian
> Peter Freddolino wrote:
>> Hi Sebastien,
>> I wouldn't worry about the "large-restraint" -> "fixed" transition; you
>> can calculate the free energy change for it, and it will be relatively
>> easy to do so because the conformations that you see are so similar. I
>> would be more worried about the zero restraint -> light restraint step,
>> for whatever the first restraint that you are using is, since to get
>> correct results you would have to sample a prohibitively large variety
>> of conformations (all populated relative interactions of the monomers in
>> your structure and associated conformations of the environment) in the
>> zero constraint state if you cannot apply some additional constraint to
>> the region of phase space that you are considering. See Journal of
>> Chemical Physics, 129, 134102 for something conceptually similar, but
>> note that in that reference the unrestrained endpoint is defined in a
>> way that vastly reduces the variety of conformations being considered.
>> You should also note two other things:
>> -the free energy changes associated with fixing the environment will be
>> massive compared to what you get from your FEP, and you may well not be
>> able to get any useful data out because the uncertainty of the
>> constraint free energy will be so large
>> -the rate-limiting step in MD calculations is usually PME, which will
>> not be accelerated by fixing a good chunk of the system, so your
>> computational savings may not be as large as you expect
>> Overall I would expect that while in principal you could follow the path
>> that you are proposing, in practice the computer time required to
>> calculate the free energy differences associated with applying the
>> constraint to a sufficient degree of accuracy will be much larger than
>> the savings you can expect from it, and you will in the process end up
>> with more statistical uncertainty in your results.
>> Best,
>> Peter
>> Sebastian Stolzenberg wrote:
>>> Dear Chris, Dear All,
>>> let me rephrase my question:
>>> I have an oligomeric structure and I can argue that FEPs on a single
>>> monomer have no significant conformational influence on the other
>>> monomers. Therefore, fixing the monomer's environment (the other
>>> monomers and the surrounding lipid/solvent molecules) saves me CPU time.
>>> This is the plan for my FEP calculation:
>>> a) step-wise *in*crease positional constraints on the monomer's
>>> environment up to a *large* constraints force constant (by manually
>>> calculating PMFs from the output of MD with the regular
>>> "constraintScaling" commands).
>>> b) NAMD-FEP on the monomer with *fixed* environment
>>> c) gradually *de*crease positional constraints on the monomer's
>>> environment down from a *large* constraints force constant (same way as
>>> in (a) )
>>> my question:
>>> I worry about the transition "large-constraints"->"fixed" environment
>>> for obtaining reasonable free energy difference. Are my concerns
>>> practically justified? If yes, how can I circumvent them to still save
>>> computational time?
>>> Thank you very much,
>>> Sebastian
>>>>> Chris Harrison wrote:
>>>>>> Sebastian,
>>>>>> Simple question first: Can NAMD do this? Yes. There's nothing in
>>>>>> the code that prevents it. Can't think of a reason NAMD wouldn't
>>>>>> execute successfully.
>>>>>> Hard question: Should you do this? During any alchemical
>>>>>> perturbation there is the possibility that the environment
>>>>>> dynamically responds by rearranging its conformation. If you can
>>>>>> justify that any changes of the conformational ensemble that occur
>>>>>> during a restrained R->A perturbation are not significantly
>>>>>> different from conformational ensemble changes that occur during an
>>>>>> unrestrained R->A perturbation, then you may be able to do this.
>>>>>> So, the atoms beyond your 30 Ang radius would have to fullfill the
>>>>>> criteria that the dynamics of those atoms you wish to restrain do
>>>>>> not respond to the R->A perturbation and the unrestrained atoms'
>>>>>> dynamical response(s) to the perturbation is not altered by the
>>>>>> presence of the restrained atoms. Please let me know how it goes.
>>>>>> I'm interested to know if it works successfully {meaning a) no NAMD
>>>>>> crashes & b) you get a correct/reasonable result}.
>>>>>> C.
>>>>>> --
>>>>>> Chris Harrison, Ph.D.
>>>>>> Theoretical and Computational Biophysics Group
>>>>>> NIH Resource for Macromolecular Modeling and Bioinformatics
>>>>>> Beckman Institute for Advanced Science and Technology
>>>>>> University of Illinois, 405 N. Mathews Ave., Urbana, IL 61801
>>>>>> Voice: 217-244-1733
>>>>>> Fax: 217-244-6078
>>>>>> Sebastian Stolzenberg <> writes:
>>>>>>> Date: Thu, 12 Mar 2009 19:50:32 -0400
>>>>>>> From: Sebastian Stolzenberg <>
>>>>>>> To:
>>>>>>> Subject: namd-l: FEP with fixed explicit environment?
>>>>>>> Return-Path:
>>>>>>> Message-ID: <>
>>>>>>> X-Spam-Status: No, score=-2.2 required=5.0 tests=AWL,BAYES_00
>>>>>>> autolearn=unavailable version=3.1.7-0+tcb1
>>>>>>> Dear Everybody,
>>>>>>> I have an equilibrated NPT structure of a protein in explicit
>>>>>>> lipid/solvent with periodic boundary conditions. Let's assume I do
>>>>>>> a mutation R105A with dual-topology FEP. Certainly, I will also
>>>>>>> need to transform a bulk water molecule (WAT) into a sodium (SOD)
>>>>>>> to keep the net charge=0. To get the final free energy difference,
>>>>>>> I will subtract delta_G(WAT->SOD) that I get from a separate run.
>>>>>>> The system is large, I was thinking about fixing all atoms of the
>>>>>>> system except for the ones around ~30A of the R105A mutation and
>>>>>>> the WAT->SOD transformation. (Of course, I will not have a
>>>>>>> fixation boundary crossing covalent bonds that leads e.g. to
>>>>>>> RATTLE constraint violations).
>>>>>>> Is this feasible with NAMD-FEP? Any troubles with periodic
>>>>>>> boundary conditions? I know that all could be locally done with
>>>>>>> implicit lipid/solvent in CHARMM, which I would like to avoid for
>>>>>>> now.
>>>>>>> Thanks so much,
>>>>>>> Sebastian

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:30 CST