They are two completely different methods for calculating vibrational
entropies. One relies more or less on PCA vectors to describe vibrations
(with the corresponding eigenvalues as the vibrational frequencies). The
other relies on a normal mode approximation whereby the Hessian is
diagonalized to obtain frequencies.
To illustrate how these methods could give drastically different entropies,
imagine a large free energy well whose fine-grained structure has many
local minima (the best example I could find in 3 seconds is here:
http://ej.iop.org/images/1478-3975/8/3/035003/Full/pb373205fig03.jpg). As
long as the barriers between the minima are relatively low, you will get
vibrational frequencies that look like they come from the vibrational
frequencies derived from the large, single free energy well. Nmode
entropies, on the other hand, will appear as though they come directly from
whatever local free energy minimum they were minimized to (so a local
minima, rather than the global well). I think these would give quite
different answers, and would help explain why quasi-harmonic calculations
take so many frames to converge.
Also, the solvation method you used for the normal mode analysis is almost
certainly different than what you used to propagate the dynamics. Ergo,
even if your system stayed in the same free energy well in a single
simulation, the shape of that well would be different between vacuum/igb=1
(the only two options for nmode) and that of the original simulation
(likely explicit solvent). As a result, you should expect different
answers.
If you want to get the *same* answers, run a simulation in implicit solvent
(igb=1) at very low temperatures (20-50K), and do your quasi-harmonic calc
with those frames. The low temperature should force you to stay in a
single well. Then, perform a normal mode calculation on a minimized
structure (making sure to adjust the temperature in both the ptraj code and
nab/sff code to get the right values, or calculate the entropy of each mode
by hand given the frequencies). I think I recall a colleague of mine
running this exact test and finding they converged to the same answer (but
it was a long time ago and the details are long buried :))
HTH,
Jason
On Tue, Jul 17, 2012 at 7:38 AM, George Tzotzos <gtzotzos.me.com> wrote:
> I would be grateful if anyone could provide an explanation regarding the
> huge differences in total Delta S given by the two methods.
>
> Below are the results obtained for exactly the same trajectories
>
> Solvated complex topology file: 2wc6_bom_solv.prmtop
> |Complex topology file: 2wc6_bom.prmtop
> |Receptor topology file: 2wc6.prmtop
> |Ligand topology file: bom.prmtop
> |Initial mdcrd(s): prod_10ns.mdcrd
> | prod_2ns.mdcrd
> | prod_4ns.mdcrd
> | prod_6ns.mdcrd
> | prod_8ns.mdcrd
> |
> |Best guess for receptor mask: ":1-141"
> |Best guess for ligand mask: ":142"
> |Ligand residue name is "BOM"
> |
> |Calculations performed using 1000 frames.
> |NMODE calculations performed using 20 frames.
>
> DELTA S total= -21.7086 +/- 1.5570
>
> ENTROPY RESULTS (QUASI-HARMONIC APPROXIMATION) CALCULATED WITH PTRAJ:
>
> DELTA S: -12.5924 -10.2513 -56.3566 -79.2006
>
>
> Many thanks in advance
>
> George
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Jul 17 2012 - 06:00:02 PDT
