Re: new NAMD user ABF setup question

From: Felipe Merino (
Date: Tue Jun 01 2010 - 20:15:53 CDT


I have a question in this regard. A couple of weeks ago in our lab we
were talking about the same topic since a PhD student here wanted to
simulate the effect of some interface mutations in some dimers, Since
these are obligated dimers, the alchemical way had the problem of not
having an adequate model for the free monomers (since on their own they
are essentially unfolded - i guess one could try to obtain a model
anyways using something like replica exchange, but it would be way too
expensive in terms of computation time). So, finally we concluded that
no PMF calculation would give good results in a reasonable time and we
arrived at the conclusion that an SMD simulation could be appropriate
for at least a qualitative result. However, isn't this approximation
considering that most of the difference in binding free energy comes
from enthalpic contributions? I ask because i guess this is not trivial
for a system as big as a protein when, even for small moelcules
sometimes the binding free energy is considerably determined by entropy.

Best regards


Jérôme Hénin escribió:
> Hi Elio,
> My guess is that a COM-COM distance will not work well for
> peptide/protein binding for any other method than irreversible
> unbinding by SMD (which will not give you binding free energies except
> on a qualitative level for comparison purposes).
> In general, binding free energies are best calculated by alchemical
> means (see the alchemical free energy tutorials).
> Jerome
> On 28 May 2010 16:03, Giacomo Fiorin <> wrote:
>> Not exactly. ABF and umbrella sampling can be applied in theory on the very
>> same (reaction) coordinates. The reason why you needed to define a specific
>> direction for pulling apart the two proteins was (I think) because of the
>> way the Gromacs implementation was written, not the general umbrella
>> sampling method.
>> In the ABF implementation in NAMD, you can try first the COM-COM distance,
>> and maybe switch later to more sophisticated ones if you think there is the
>> need for. Also, since it's implemented within the colvars module, you can
>> even reuse the exact same variables/coordinates with ABF and umbrella
>> sampling, and be able to compare their results, if that's what you are
>> interested in.
>> Giacomo
>> ---- ----
>> Dr. Giacomo Fiorin
>> ICMS - Institute for Computational Molecular Science - Temple University
>> 1900 N 12 th Street, Philadelphia, PA 19122
>> ---- ----
>> On Fri, May 28, 2010 at 8:48 AM, Elio Cino <> wrote:
>>> Thanks Giacomo. Very helpful reply. I think that I am treating the ABF
>>> setup too much like that for umbrella sampling where a specific coordinate
>>> is define. For ABF if I simply specify a range of COMdistance between the
>>> protein and ligand then I presume that conformations in this defined COM
>>> range will eventually be sampled. This way I do not need to pull them apart
>>> along a straight line, for example. Hopefully I have this straight. Ill have
>>> to read a bit more on ABF to be sure. Thanks again.
>>> ----- Original Message -----
>>> From: Giacomo Fiorin <>
>>> Date: Thursday, May 27, 2010 5:27 pm
>>> Subject: Re: namd-l: new NAMD user ABF setup question
>>> To: Elio Cino <>
>>> Cc:
>>>> Hi Elio, the most intuitive collective variable, or reaction coordinate
>>>> is the COM-COM distance, whichever its direction may be. If you want to
>>>> pull them apart (and back together!) along a specific direction, you have to
>>>> restrict the variable to be on the projection along that "specific
>>>> coordinate" as you said: do you mean Cartesian coordinate (x, y or z)? My
>>>> suggestion is to use distanceZ from the ABF implementation in NAMD 2.7b1 and
>>>> 2.7b2, and also define distanceXY with a restraint to be zero, so that
>>>> you'll have a straight line, if that's what you want.
>>>> The comment you were referring to sounds specific to a well defined
>>>> system, which may not be similar to yours. The boundaries defining the
>>>> relevant region of your system are your choice entirely.
>>>> You should check out the ABF paper and compare it with umbrella sampling
>>>> using your own intuition. Too often comparisons between free energy methods
>>>> are based on specific examples that may not necessarily be applicable to
>>>> your case.
>>>> In any case, when the variables are appropriately chosen, ABF DOES give
>>>> you an accurate PMF within the range (boundaries) you defined.
>>>> Giacomo
>>>> ---- ----
>>>> Dr. Giacomo Fiorin
>>>> ICMS - Institute for Computational Molecular Science - Temple
>>>> University
>>>> 1900 N 12 th Street, Philadelphia, PA 19122
>>>> ---- ----
>>>> On Thu, May 27, 2010 at 3:10 PM, Elio Cino <> wrote:
>>>>> Hello. I have been trying to calculate free energies of peptide binding
>>>>> to a protein with umbrella sampling in gromacs with mixed results. Anyways,
>>>>> I was hoping the ABF method may help me get accurate results faster than
>>>>> umbrella sampling. I have read some of the ABF tutorial and manual, but am a
>>>>> bit confused about the ABF setup. I set the boundaries for the COM distance
>>>>> between the peptide and protein, but I think I need to specify a direction
>>>>> also since I want to pull the peptide away from the protein along a specific
>>>>> coordinate. I am unsure if they is correct, and if so how to specify the
>>>>> direction. Also, I read a comment saying that much of the ABF calculation
>>>>> time is spent climbing uphill to overcome the forces (between peptide and
>>>>> protein) at low COM distances and to avoid this, the lower boundary should
>>>>> be increased. Will this still provide an accurate PMF? It seems that
>>>>> excluding these low boundary states would cause some errors in the delta G.
>>>>> Any help is appreciated. Thanks.
>>>>> Elio Cino
>>> Elio Cino

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:55:50 CST