[MMTK] modeling arbitrary structures

DeLano, Warren warren@sunesis.com
Tue, 7 Aug 2001 11:27:50 -0700


>=20
> > > 	As the author of PyMOL, I am very interested in this issue.  I
> > > currently do a substantial amount of modeling of=20
> arbitrary structures
> > > using PyMOL with batchmin (commercial) as the back-end. =20
> It would be
> > > great if MMTK could be modified to take arbitrary=20
> geometries straight
> > > out of PyMOL in the same way.
>=20
> What information would be available from PyMOL?

PyMOL has a minimalist molecular representation -- essentially PDB
records with the following additional properties for purposes of
modeling:

formal_charge (for MOL files)
partial_charge (for calculations)
numeric_type (for MMFF)
text_type (for Amber)
32 flags (for indicating restrained atoms, etc.)

there are some other atom properties not related to modeling...

There is also a bond list, with each bond having unit valence but no
implicit aromaticity at present.

The classes are defined in modules/chempy/__init__.py and
modules/chempy/models.py

If additional information is need it could be introduced...

Assigning MMFF atom types is quite trivial (just pattern perception),
but doing the same for Amber would be quite a challenge.  Although I've
shoe-horned a number of organics into Amber in order to answer some
basic questions about geometry in a protein context, accurate parameters
are quite system-specific and require substantial effort to generate. =20

	Furthermore, I question whether a simplistic forcefield such as
Amber (just bonds, angles, torsions, impropers, and pairwise nonbonded
potentials) is sufficient to capture dynamics (or even geometry) of
small molecules in a general sense.  At the very least, to do this right
you would basically need to start from scratch with a new set of generic
atom types for small molecules and develop "compromise" parameters for
these atom types.  That is a Ph.D. thesis sized project, not something
one could cook up on a weekend...

- Warren