Application of NMR and Molecular Docking in Structure-Based Drug Discovery - NMR of Proteins and Small Biomolecules - page 8

Many computer programs are available to perform or assist with molecular
docking. The vast number of docking programs makes it impractical to describe
them all in detail within a single review (for other reviews please see [ 85 - 89 ]). Each
docking program does have some unique features that make them particularly
useful for a given situation or problem. However, nearly all the docking programs
consist of two primary components: docking (or searching) and scoring [ 30 , 31 ].
Docking refers to the sampling of the ligand's conformation space and its orienta-
tion relative to a receptor. Scoring is used to evaluate and rank the current pose of
the ligand.

3.1 Docking

The docking process requires, at a minimum, two inputs: the three-dimensional
structures of the receptor (protein) and the ligand. The most common simplification
to the docking process is to keep the structure of the receptor rigid and stationary.
Only the ligand is typically allowed to be flexible as it is docked to the protein.
Keeping the protein rigid significantly minimizes the complexity of the calculation.
Sampling the conformations and orientations of the ligand is done using systematic
or stochastic methods [ 30 , 31 ].
Systematic search methods attempt to sample all of the possible conformations
of a ligand by incrementing the torsional angles of each rotatable bond. Unfortu-
nately, this technique is computationally expensive due to the exponential increase
in the number of possible conformations ( N conf ) as the number of rotatable bonds
increases:

Y Y

N n inc

360
y i;j ;

N conf ¼ (2)

i ¼ 1 j ¼ 1

where N represents the number of rotatable bonds, n inc is the number of incremental
rotations for each rotatable bond, and y i,j is the size of the incremental rotation for
each rotatable bond. As a result, purely brute force systematic approaches are
generally not used. Instead, most systematic searches require the use of efficient
shortcuts. As an illustration, MOLSDOCK [ 90 ] uses mutually orthogonal Latin
squares (MOLS) to identify optimal ligand conformations. Latin squares are an
N N matrix, where each parameter (torsion angle value) occurs only once in each

row and column. Orthogonal Latin squares are two or more superimposed N N

matrices, where each parameter still only occurs once in each row and column.
MOLS are used to identify the N 2 subset of ligand conformations used to calculate
binding energies. Simply, only a small subset of the possible ligand conformations
is sampled to construct the potential surface and identify the minima.
Perhaps the most commonly utilized systematic search method is incremental
construction, which is used by DOCK [ 41 ], FlexX [ 42 ], E-Novo [ 91 ], LUDI
[ 45 , 46 ], ADAM [ 92 ], and TrixX [ 93 ]. In this particular method, the ligand is