Small Molecule Docking from Theoretical Structural Models - Computational Modeling of Biological Systems - page 78

energy terms precomputed in a regular grid. 2 Scoring is performed considering
ligand-entropic terms and desolvation contributions in addition to ligand-protein
interaction terms. GOLD [ 21 ], another very popular program, uses a sampling
protocol similar to the genetic algorithm implemented in AUTODOCK and a very
wide range of well-validated scoring functions, which include specific corrections
such as those for metal ions and covalent interactions [ 22 ]. This program includes
also specific scoring functions for kinases and offers the possibility to incorporate
user-refined scoring functions. The program FLExX [ 23 ], which has also an ex-
cellent record of success, uses a geometry-fitting algorithm derived from computer
vision engineering, where drugs grow in optimum orientations and conformations
at the binding site from an original seed fragment. The program permits the
introduction of knowledge-based pharmacological restraints and the incorporation
of essential water molecules and crucial metal ions in the binding site. Scoring
is based on a simple physical scoring function based on OPLS [ 24 ] force-field
parameters. ICM [ 25 ], a powerful program to fit small ligands to proteins, uses
a smoothed atomistic energy function coupled with a Monte Carlo algorithm in
internal coordinates to sample the drug-protein binding space. Its scoring function
contains the usual contributions plus two desolvation correction terms. GLIDE [ 26 ],
a widely used docking program in the pharmaceutical industry, uses a “funnel
strategy” where each pose passes a series of hierarchical filters that evaluate the
ligand-receptor interactions, including spatial fit, complementarity of interactions
using a grid-based method, and finally an evaluation and minimization using OPLS-
AA nonbonded ligand-receptor interaction energy. GLIDE incorporates a variety
of scoring functions with increasing computational complexity. MedusaDock [ 27 ],
a recently developed software, is a docking method which models both ligand and
receptor flexibility in a rapid manner by using sets of discrete rotamers, obtaining
quite good results with targets which are known to be very flexible.
In addition to those implemented in standard programs, many other scoring
functions have been developed (for a review see [ 28 ]), using experimentally
calibrated master equations similar to that in (1).

D ˛E ele C ˇE vW C E HBond C ıG desolv C "S lig C E lig G binding

dist C 'G others ; (1)

where E ele and E vW stand for usual electrostatic and van der Waals terms—typically
smoothed to avoid nuclei discontinuities. Hydrogen bonds contribution is sometimes
explicitly included in E Hbond , while in others it is captured by E ele and E vW .The
ligand and protein desolvation contribution (typically computed from occluded
surface/volumes) are included in G desolv , the loss of ligand entropy upon binding
is introduced in S lig (typically roughly approximated by counting the number of
rotable bonds in the ligand), and the constrained energy is captured by E lig .Other
additional terms can be included, such as corrections for covalent interactions,

2 Representation of the receptor energetic contributions (mainly electrostatic and van der Waals) to
be read during the ligand scoring.