Chemistry Reference
In-Depth Information
interactions between aromatic rings; metal atoms and effective
number of rotatable bonds in the ligand.
Many protein-ligand programs use different empirical energy terms in
their empirical scoring functions. Training sets may be necessary to
avoid problems such as double counting.
c)
Knowledge-based scoring functions are also known as statistical-
potential based scoring functions. Here energy potentials (can be
derived with experimental information) are used. Using the inverse
Boltzmann relation, pairwise potentials can be directly obtained from
the occurrence frequency of atom pairs. In the simple fluid system, by
the inverse Boltzmann method, the interaction potentials are the mean-
force potentials. For complex protein systems, the Boltzmann method
can convert a histogram of atom-atom distances into a suitable
function with the dimensions of energy.
The knowledge-based scoring function potentials are extracted from structures to
reproduce the known affinities by fitting. They are robust and quite insensitive to
training sets. In addition, the scoring process can be as fast as using empirical
scoring functions. The disadvantage is the challenge of reference states. Often the
reference state is approximated by an atom-randomized state by ignoring effects
such as excluded volume and interatomic connectivity. Some scoring functions
consists of a distance-dependent pair-potential term and a surface-dependent
single potential term. Corrections or scaling can also be introduced to improve the
accuracy of the reference states. The potential of mean force (PMF) is one of the
first knowledge-based scoring functions extensively tested for affinity [547].
The problem of the reference state is a problem using either more physical
approximations or atom-randomized states. The pairwise potentials are not
sensitive enough to ligand positions highlighting the problem for virtual screening
and binding mode predictions. Consequently, a knowledge-based potential scoring
function was developed (ITScore) in which the requirement of accurate reference
state calculations is avoided [548]. The pair potential can be iteratively adjusted to
reproduce experimental training set distribution functions, which are able to
