[MMTK] Forcefields

Paul D Soper Paul.D.Soper@USA.dupont.com
Tue, 7 Aug 2001 07:53:18 -0400

As Konrad pointed out, the key issue in doing useful things for
non-biological molecules in MMTK is how to compute energies.  I believe
there are two generic approaches one can use: 1) create AMBER parameters as
needed or 2) build new forcefield capability in MMTK.

Creating AMBER parameters is relatively straightforward.  There are a
number of forcefields whose diagonal terms (bond stretch, angle bend,
dihedral torsion, van der Waals, electrostatic) could reasonably be
expected to be in the right ballpark for AMBER.  This can be checked by
looking at molecules which both AMBER and the other forcefield can handle
and comparing the force constants.  If they are close, just use the numbers
from the other forcefield in AMBER.  Partial atomic charges are easily
calculated from Gaussian output using the constrained ESP fitting code
included in MMTK.  (Subject to all the usual caveats about partial atomic
charges - conformational dependence, etc.)  This method is easiest to apply
when the atom types exist in AMBER and only the force constants are
missing, but it is also possible to define new atom types.

The other alternative would be to port a forcefield to MMTK which is
designed to be more generic than AMBER.  The forcefield should cover more
atom types than AMBER, include many more force constants, and include rules
for estimation of force constants when they are not known.  My suggestion
would be the "Class II" CFF and related forcefields from
Biosym->MSI->Accelrys.  The port would not be trivial since the forcefields
abound in cross terms which  are absent in AMBER (bond stretch/angle bend,
bond stretch/dihedral torsion, etc.).  On the other hand, these forcefields
use rules to assign partial atomic charges, so the ESP fitting step is
eliminated.  If some brave soul is actually trying to do this I hope he or
she will speak up.  (Alas, I've only thought about doing it.)

- Paul