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

From: Holger Gohlke <gohlke.scripps.edu>

Date: Thu, 20 Feb 2003 10:35:57 -0800

Hi,

*> Dear ambers,
*

*> I have some doubts regarding the usage of Nmode. I
*

*> want to compare the free energy of formation of two
*

*> DNA systems (1437 atoms). For that, I have done
*

*> steepest descent and conjugate gradient minimiztaion
*

*> using sander_classic and I am using Nmode for
*

*> Newton-Rapson minimization. I have following doubts
*

*> regarding this:
*

*> (i) If I have carried out Newton-Rapson minimization
*

*> for the convergence of 0.0001, convergence is not
*

*> reached even after 18 hours. So, I have changed the
*

*> convergence to 0.01 and now I am able to reach the
*

*> convergence within 9 hours. If I use the convergence
*

*> of 0.01 will it affect the calculation of free-energy?
*

This behavior of the Newton-Raphson minimization is not unexpected

because (in principle) in each step of the minimization the Hessian

matrix of the system has to be generated and diagonalized. And this is

rather time-consuming for systems of your size (calculating the Hessian

scales with the square of the number of atoms; diagonalizing it scales

with the cube). In the realm of mm_pbsa, we use the conjugate-gradient

method of sander instead to reach gradients below 0.0001 kcal/(mol A).

*> (ii) I have used nvec=200, bdwnhl=0.01 and dfpred=0.1
*

*> as given in the example file (amber 4.1). Can I use
*

*> these values? otherwise how to choose these values.
*

If you don't use the NR minimization any more, bdwnhl and dfpred will be

obsolete. The choice of the nvect(?) parameter depends on what you want

to do. If you just want to use Nmode to calculate entropies, you can set

nvect=0. (I hope, nmode of amber4.1 and of amber7 behaves similar in

this point.)

*> (iii) In the manual (Nmode) it is mentioned that
*

*> dielc=1 and idiel=0 should be used for r^^2 distance
*

*> dependence. But, for minimization (using
*

*> sander_classic and also for anal) I have used dielc=4
*

*> and idiel=0 for 4rij. My question is that which one
*

*> should be used for the calculations - r^^2 distance
*

*> dependence or 4rij.(I have used scee=1.2 for all the
*

*> calculations, since I am using parm98.dat force field)
*

I think it is appropriate to use a distance-dependent dielectric

constant of 4r for minimization and then also for NM analysis. So this

would result in a "4*r^^2" distance dependence.

*> (iv) I have done dynamics for 4ns. Is it sufficient
*

*> if I take the last 1.5ns (at 20ps intervals) alone for
*

*> the calculation?
*

Difficult to say in advance. But you will see how large the standard

error of the mean is. In view of the rather time-consuming calculations,

you might also consider using snapshots at larger time intervals first.

*> (v) For the calculation of free energy I am using
*

*> MM-GBSA. In /amber6/src/mm_pbsa/SOURCS/GB.F file only
*

*> the solvent dielectric constant (eps=80) is given and
*

*> not the solute dielectric constant (=1). Where should
*

*> it be given?
*

If I remember right, this GB model has been parametrized for eps(int)=1

which is considered implicitly in the code.

Best regards

Holger

*> Kindly, tell your suggestions regarding this.
*

*> Thank you very much.
*

*>
*

*> Thenmalar
*

*>
*

*> ________________________________________________________________________
*

*> Missed your favourite TV serial last night? Try the new, Yahoo! TV.
*

*> visit http://in.tv.yahoo.com
*

Date: Thu, 20 Feb 2003 10:35:57 -0800

Hi,

This behavior of the Newton-Raphson minimization is not unexpected

because (in principle) in each step of the minimization the Hessian

matrix of the system has to be generated and diagonalized. And this is

rather time-consuming for systems of your size (calculating the Hessian

scales with the square of the number of atoms; diagonalizing it scales

with the cube). In the realm of mm_pbsa, we use the conjugate-gradient

method of sander instead to reach gradients below 0.0001 kcal/(mol A).

If you don't use the NR minimization any more, bdwnhl and dfpred will be

obsolete. The choice of the nvect(?) parameter depends on what you want

to do. If you just want to use Nmode to calculate entropies, you can set

nvect=0. (I hope, nmode of amber4.1 and of amber7 behaves similar in

this point.)

I think it is appropriate to use a distance-dependent dielectric

constant of 4r for minimization and then also for NM analysis. So this

would result in a "4*r^^2" distance dependence.

Difficult to say in advance. But you will see how large the standard

error of the mean is. In view of the rather time-consuming calculations,

you might also consider using snapshots at larger time intervals first.

If I remember right, this GB model has been parametrized for eps(int)=1

which is considered implicitly in the code.

Best regards

Holger

-- +++++++++++++++++++++++++++++++++++++++++++++ Dr. Holger Gohlke Dept. of Molecular Biology, TPC15 The Scripps Research Institute 10550 N. Torrey Pines Rd. La Jolla CA 92037 USA phone: +1-858-784-9788 fax: +1-858-784-8896 email: gohlke.scripps.edu +++++++++++++++++++++++++++++++++++++++++++++Received on Thu Feb 20 2003 - 18:53:00 PST

Custom Search