- Contemporary messages sorted: [ by date ] [ by thread ] [ by subject ] [ by author ] [ by messages with attachments ]

From: Carlos Simmerling <carlos.simmerling.gmail.com>

Date: Wed, 2 Feb 2011 10:45:21 -0500

2D umbrella sampling can be very difficult to do well. I suggest that

when you compare the surfaces, you must do so in the context of the

uncertainty in each surface as calculated by fully independent

simulations. You must also carefully analyze overlap in distributions

between windows- not only in RC1 and RC2, but in other measures as

well. For complex systems, PMF calculation requires very detailed

analysis before one can have any confidence in the PMFS.

On Wed, Feb 2, 2011 at 10:40 AM, dhacademic <dhacademic.gmail.com> wrote:

*> Thanks for the reply. Let me make it more clear.
*

*>
*

*> I have one FE-surface and two FE-profiles:
*

*> (a) 2D-FES (umbrella sampling with both RC1 and RC2);
*

*> (b) quasi-1D-FE-profile (PMF calculation with varied RC1 and fixed RC2 (for
*

*> example, RC2=1). So the result is not integrated over the RC2. In reality,
*

*> it should be a thin slice of 2D-results);
*

*> (c) 1D-FES (umbrella sampling only with RC1).
*

*>
*

*> Results from (b) and (c) are consistent. (b) and (a) are inconsistent, which
*

*> makes me confused. I can understand that 2D-FES may differ to "quasi-1D"
*

*> FE-profile because of the integral over RC2 dimension. But in present case,
*

*> my quasi-1D FE-profile is only one slice of 2D-FE-surface, which should not
*

*> change the location of minimum.
*

*>
*

*> I will check the parameters as well as simulations again. But any other
*

*> suggestion is appreciated!
*

*>
*

*> Best,
*

*> Hao
*

*>
*

*>
*

*> On Tue, Feb 1, 2011 at 9:49 PM, Daniel Sindhikara <sindhikara.gmail.com>wrote:
*

*>
*

*>> Not sure if I understand your situation but it seems you are trying to
*

*>> interpret the
*

*>> difference between a 2d and 1d FES? You call the 1D a "quasi-1D" because
*

*>> you
*

*>> integrated over the 2nd dimension, but the 2D surface integrates over all
*

*>> other dimensions.
*

*>> If your 1D shows a minimum at one end and the 2D shows it at the other end
*

*>> on the same axis, then the discrepancy is clearly entropy.
*

*>> In terms of your coordinates, RC1=10,RC2=1 is the lowest free energy,
*

*>> but there are many low free energy points at RC1=1, which is why the
*

*>> integral over RC2 shows this to be a minimum.
*

*>>
*

*>> This is all assuming that your calculations were correct, and that your
*

*>> umbrella sampling simulations were well converged, and your force-field
*

*>> accurate (a big heap of assumptions).
*

*>>
*

*>> If this result is contrary to what you expect, may I suggest you recheck
*

*>> your simulation parameters.
*

*>>
*

*>>
*

*>> --Dan
*

*>>
*

*>> On Wed, Feb 2, 2011 at 5:55 AM, dhacademic <dhacademic.gmail.com> wrote:
*

*>>
*

*>> > Hi everyone,
*

*>> >
*

*>> >
*

*>> > I have a problem in 2D PMF calculations.
*

*>> >
*

*>> >
*

*>> >
*

*>> > There are two reaction coordinates (RC) in my system. RC1 stands for
*

*>> > conformational change of protein (RC1 ranges from 1 to 10, where RC1=1
*

*>> > means
*

*>> > closed state of protein, RC1=10 means open state), and RC2 is the
*

*>> distance
*

*>> > between ligand and active site (RC2 ranges from 1 to 10, where RC2=1
*

*>> means
*

*>> > ligand bound state, RC2=10 means ligand escaped state). After the
*

*>> > time-consuming umbrella sampling calculations (with amber9) on two RCs,
*

*>> the
*

*>> > 2D free energy profile can be obtained with WHAM program. However, the
*

*>> open
*

*>> > state of protein with ligand bound to active site (RC1=10, RC2=1) is
*

*>> found
*

*>> > to be a minimum on the free energy surface, while the closed state with
*

*>> > ligand bound has higher energy (RC1=1, RC2=1). This is opposite to what I
*

*>> > have expected, where the closed state of protein with ligand bound should
*

*>> > be
*

*>> > a stable state.
*

*>> >
*

*>> >
*

*>> >
*

*>> > Then the umbrella sampling data of different RC1 (from 1 to 10) with
*

*>> fixed
*

*>> > RC2 (RC2=1) are used to get quasi-1D free energy profile of protein
*

*>> > conformational change with ligand bound, and the results seem to be
*

*>> > reasonable: the open state is energetic unfavorable, and the closed state
*

*>> > has a minimum. The system stability in quasi-1D free energy profile is
*

*>> > reasonable.
*

*>> >
*

*>> >
*

*>> >
*

*>> > Besides, umbrella sampling on RC1 (no bias potential is imposed on RC2,
*

*>> and
*

*>> > the ligand position and orientation is almost unchanged because there are
*

*>> > lots of favorable interactions between ligand and active site is strong)
*

*>> > was
*

*>> > done on the same system. The generated 1D-PMF results is quite similar to
*

*>> > the quasi-1D one, and quite different to the 2D-PMF results when RC2=1.
*

*>> >
*

*>> >
*

*>> >
*

*>> > All the structures as well as the data have been carefully checked.
*

*>> > Everything looks good, but the results look strange. I do not understand
*

*>> > why
*

*>> > the stability of the system in 2D-PMF and quasi-1D-PMF is different. As
*

*>> the
*

*>> > results from 2D-PMF are inconsistent with common knowledge of
*

*>> > protein-ligand
*

*>> > complex, I think there may be something wrong with my 2D-PMF
*

*>> calculations.
*

*>> > Can anyone give some suggestions? Thanks in advance!
*

*>> >
*

*>> >
*

*>> >
*

*>> > Best,
*

*>> >
*

*>> > Hao
*

*>> > _______________________________________________
*

*>> > AMBER mailing list
*

*>> > AMBER.ambermd.org
*

*>> > http://lists.ambermd.org/mailman/listinfo/amber
*

*>> >
*

*>>
*

*>>
*

*>>
*

*>> --
*

*>> Dr. Daniel J. Sindhikara
*

*>> Institute for Molecular Science
*

*>> E-mail: sindhikara.gmail.com
*

*>> Website: http://sites.google.com/site/dansindhikara/
*

*>> --
*

*>> _______________________________________________
*

*>> AMBER mailing list
*

*>> AMBER.ambermd.org
*

*>> http://lists.ambermd.org/mailman/listinfo/amber
*

*>>
*

*> _______________________________________________
*

*> AMBER mailing list
*

*> AMBER.ambermd.org
*

*> http://lists.ambermd.org/mailman/listinfo/amber
*

*>
*

_______________________________________________

AMBER mailing list

AMBER.ambermd.org

http://lists.ambermd.org/mailman/listinfo/amber

Received on Wed Feb 02 2011 - 08:00:10 PST

Date: Wed, 2 Feb 2011 10:45:21 -0500

2D umbrella sampling can be very difficult to do well. I suggest that

when you compare the surfaces, you must do so in the context of the

uncertainty in each surface as calculated by fully independent

simulations. You must also carefully analyze overlap in distributions

between windows- not only in RC1 and RC2, but in other measures as

well. For complex systems, PMF calculation requires very detailed

analysis before one can have any confidence in the PMFS.

On Wed, Feb 2, 2011 at 10:40 AM, dhacademic <dhacademic.gmail.com> wrote:

_______________________________________________

AMBER mailing list

AMBER.ambermd.org

http://lists.ambermd.org/mailman/listinfo/amber

Received on Wed Feb 02 2011 - 08:00:10 PST

Custom Search