5.1.4 Pattern Identification and Scoring
Once the features extracted for each ligand in the dataset, a pattern is identified as a
set of relative positions in the 3D space, each linked to a feature. If a ligand holds a
set of features in at least one of its conformations, the set of features can be aligned
with the corresponding locations. Most of the methods are based on spatially
overlaying conformations of various compounds with the pharmacophores points
with minimal root mean square alignment errors. One can roughly classify the
alignment methods as either point or property-based. In the first class of algorithms,
pairs of pharmacophoric features are generally aligned using a least-squares fitting
using clique detection methods [76, 77]. According to the graph-theoretical
approach to molecular structures, a clique is a maximum completely connected
sub-graph, which recognizes all imaginable combinations of atoms/functional
groups to find out common substructures for the alignment. Property-based or
field-based algorithms utilize grid or field descriptors, based on molecular properties such as volume, shape, charge distribution, electron density and electrostatic
potentials of molecules. A 3D grid is generated about a ligand by computing the
interaction energy components between the ligand and a probe placed at each grid
point. Properties are calculated on a grid and later converted to a set of Gaussian
representations. A number of either random or thoroughly sampled initial configurations are then generated followed by local optimizations with some similarity
measure of the intermolecular overlap of the Gaussians.
After obtaining the pharmacophore candidates in the previous stages, they are
generally scored and ranked. The basic obligation of a scoring scheme is implemented such that a high score implies higher chance of the ligands mapping to the
pharmacophore model. Despite the great advances, molecular alignment handling
ligand flexibility and proper selection of training set compounds are considered as
the biggest challenges in ligand-based pharmacophore modelling.
5.2 Structure-Based Pharmacophore Model Generation
Structure-based pharmacophore modelling requires the 3D structure of the receptor
or a receptor–ligand complex. The models are generated based on the spatial
relationships of complementary interaction features of the binding pockets followed
by selection and assembly of features to generate pharmacophore models.
5.2.1 Active Site Identification
The input for receptor-based pharmacophore modelling is the three-dimensional
structure of a receptor usually in PDB format. The receptor binding pocket is
identified using a spherical probe with customizable radius and location to include
the binding site as well as the key interacting residues involved with ligands.
Pharmacophore Modelling and Screening: Concepts, Recent …
35
Once the features extracted for each ligand in the dataset, a pattern is identified as a
set of relative positions in the 3D space, each linked to a feature. If a ligand holds a
set of features in at least one of its conformations, the set of features can be aligned
with the corresponding locations. Most of the methods are based on spatially
overlaying conformations of various compounds with the pharmacophores points
with minimal root mean square alignment errors. One can roughly classify the
alignment methods as either point or property-based. In the first class of algorithms,
pairs of pharmacophoric features are generally aligned using a least-squares fitting
using clique detection methods [76, 77]. According to the graph-theoretical
approach to molecular structures, a clique is a maximum completely connected
sub-graph, which recognizes all imaginable combinations of atoms/functional
groups to find out common substructures for the alignment. Property-based or
field-based algorithms utilize grid or field descriptors, based on molecular properties such as volume, shape, charge distribution, electron density and electrostatic
potentials of molecules. A 3D grid is generated about a ligand by computing the
interaction energy components between the ligand and a probe placed at each grid
point. Properties are calculated on a grid and later converted to a set of Gaussian
representations. A number of either random or thoroughly sampled initial configurations are then generated followed by local optimizations with some similarity
measure of the intermolecular overlap of the Gaussians.
After obtaining the pharmacophore candidates in the previous stages, they are
generally scored and ranked. The basic obligation of a scoring scheme is implemented such that a high score implies higher chance of the ligands mapping to the
pharmacophore model. Despite the great advances, molecular alignment handling
ligand flexibility and proper selection of training set compounds are considered as
the biggest challenges in ligand-based pharmacophore modelling.
5.2 Structure-Based Pharmacophore Model Generation
Structure-based pharmacophore modelling requires the 3D structure of the receptor
or a receptor–ligand complex. The models are generated based on the spatial
relationships of complementary interaction features of the binding pockets followed
by selection and assembly of features to generate pharmacophore models.
5.2.1 Active Site Identification
The input for receptor-based pharmacophore modelling is the three-dimensional
structure of a receptor usually in PDB format. The receptor binding pocket is
identified using a spherical probe with customizable radius and location to include
the binding site as well as the key interacting residues involved with ligands.
Pharmacophore Modelling and Screening: Concepts, Recent …
35