molecular rearrangement 185lations may be performed at different levels of approxima-
tion that can be divided into two classes: semiempirical
and ab initio (from the beginning, i.e., based on first princi-
ples). Even so, a geometrical optimization of a molecule
composed of 30 atoms that is nearly instantaneous on a
personal computer (PC) using MM methods may require
several minutes using semiempirical QM methods and an
hour or even days using ab initio techniques.
Semiempirical QM methods (e.g., PM3, AM1, and
MNDO) employ a variety of simplifications and experimen-
tally derived elemental parameters to speed up calculations
versus ab initio methods. All implementations support most
of the elements in biologically and commercially important
organic compounds. Some programs also support a wide
range of transition metals. Semiempirical QM calculations
provide very good geometries and associated ground-state
properties: atom-centered charges, ionization potential,
heats of formation, and some indication of reactivity based
on the frontier molecular orbitals (the highest occupied and
lowest unoccupied molecular orbitals, HOMO and LUMO).
But these methods are generally not suited for studying
reaction mechanisms. The limitations of semiempirical QM
methods are offset by the ability to conduct QM calcula-
tions on systems consisting of hundreds of atoms, including
small enzymes.
The ab initio QM methods are based solely on the
laws of quantum mechanics and therefore have the broad-
est applicability. They can be carried out at different levels
of approximation in order to balance the required accu-
racy against the computational demands. The quality of
the calculations is principally determined by the selected
basis set (functions that describe the atomic orbitals) and
the treatment of electron correlation (interaction between
electrons). Generally, moderate basis sets are sufficient
for accurate ground-state calculations, but large basis
sets and proper treatment of electron correlation are
required to model excited states, transient species, or
chemical reaction mechanisms. Fortunately, modern treat-
ment of electron correlation, based on density functional
theory, has made high-quality calculations using a PC fea-
sible for systems containing tens of atoms, sufficient to
study enzyme-active sites. Applications of ab initio QM
include designing new catalysts, semiconductors, and
dyes and studying atmospheric chemistry, such as the
impact of greenhouse gases and chlorofluorocarbons (fre-
ons) on ozone depletion.
Another area of molecular modeling involves devel-
opment of quantitative structure-activity or structure-
property relationships (QSAR and QSPR). These studies
use a range of statistical methods (linear and nonlinear
regression, neural nets, clustering, genetic algorithms,
etc.) to correlate molecular properties determined experi-
mentally and derived from MM or QM calculations against
the known end-use biological activities or physical or
chemical properties for a large training set of molecules.
Such activity models can then be used to predict the per-
formance of similar molecules, even ones that do not yet
exist. A key to success of QSAR studies is that the compo-
sition and structure, i.e., chemistry, of the test compound
must be represented in the training set, otherwise the pre-
dictions can be very misleading. Even with this limitation,
it is often possible to generate hundreds or even thou-
sands of ideas that can be rapidly screened for the most
promising compounds to advance for laboratory synthesis
and testing. Such high-throughput screening (HTS) is
rapidly being adopted as standard research practice. In
particular, pharmaceutical companies employ ADME
(adsorption, digestion, metabolism, and elimination) and
TOX (toxicology) models in their screening process.
Indeed, regulatory agencies in the United States and Euro-
pean Union also employ QSAR models as part of their
review of new materials, and some groups have proposed
them as replacements for safety studies involving animals.
These same approaches are used to predict protein struc-
ture activities (proteomics) and decipher genetic codes
(genomics).
The advent of advanced computer graphics worksta-
tions during the 1990s dramatically improved the scientific
research communities’ access to molecular-modeling
capabilities. Continued advances are rapidly making com-
putational molecular modeling an integral part of chemistry
and its related scientific fields. Chemists, knowledgeable
about the available modeling tools, now have the ability to
test ideas on their PCs before stepping into the labora-
tory, thereby maximizing the likelihood of success and
eliminating unnecessary work. Chemists once sketched
molecules on paper and built molecular-scale models on
their desks. Today they assemble them on a three-dimen-
sional computer display, optimize the structure quickly,
conduct a conformational search, compute spectral prop-
erties, estimate physiochemical properties, and compute
and display molecular orbitals or space-filling models
with mapped electrostatic charges—all of which can be
dynamically rotated, resized, modified, or combined into
new models.—Karl F. Moschner, Ph.D.,is an organic
chemistry and scientific computing
consultant in Troy, New York.