Application of NMR and Molecular Docking in Structure-Based Drug Discovery - NMR of Proteins and Small Biomolecules - page 11

1
K D :

D G binding ¼ RT ln

(4)

Unfortunately, accurately calculating the binding free energy is very challenging
due to the many forces that influence binding. In molecular docking, there are five
primary types of scoring functions: force field-based, empirical, knowledge-based,
shape-based, and consensus [ 100 - 102 ].
Force field-based scoring functions [ 30 , 31 ] are used to calculate the free energy
of binding by combining the receptor-ligand interaction energy and the change in
internal energies of the ligand based on its bound conformation (Fig. 4 ). The
internal energy of the receptor is usually ignored since the receptor is kept rigid
in most docking programs. The protein-ligand binding energies are typically
defined by van der Waal forces, hydrogen bonding energies, and electrostatic
energy terms. The van der Waals and hydrogen bonding terms often utilize a
Lennard-Jones potential function, while the electrostatic terms are described by a
coulombic function. Unfortunately, these interaction energies were originally
derived from measuring enthalpic interactions in the gas phase. Of course,
receptor-ligand binding interactions actually occur in an aqueous solution, which
introduces additional interactions between the solvent molecules, the receptor, and
the ligand. Protein-ligand binding energies are also dependent on the entropic
changes that occur upon binding, which include torsional, vibrational, rotational,
and translational entropies. Most entropy and solvation-based energy terms can't be
calculated using force field-based scoring functions. As a result, force field-based
scoring functions are incomplete and inaccurate.
Empirical scoring functions [ 103 - 106 ] are similar to force field-based scoring
functions since they use a summation of individual energy terms. But empirical
scoring functions also attempt to include solvation and entropic terms. This is
typically achieved by using experimentally determined binding energies of
known ligand-receptor interactions to train the scoring system using regression
analysis. Empirical scoring functions are fast, but the accuracy is completely
dependent upon the experimental data set used to train the scoring function.
In general, empirical scoring functions are reliable for ligand-receptor complexes
that are similar to the training set.
Knowledge-based scoring functions [ 107 - 109 ] are fundamentally different from
force field-based and empirical scoring functions. Knowledge-based scoring
functions don't attempt to calculate the free energy of binding. Instead, these
scoring functions utilize a sum of protein-ligand atom pair interaction potentials
to calculate a binding affinity. The atom pair interaction potentials are generated
based upon a probability distribution of interatomic distances found in known
protein-ligand structures. The probability distributions are then converted into
distance-dependent interaction energies. In this manner, knowledge-based scoring
functions allow for the modeling of binding interactions that are not well under-
stood. The approach is also very simple, which is useful for screening large com-
pound libraries. Unfortunately, knowledge-based scoring functions are designed