reduce the conformational space. The conformers with lowest energy or representatives from clusters of similar conformers are chosen to take part in model
generation. In some other software, conformational search is parallelly performed
along with pattern identification by retaining the conformers that possess certain
features in a particular spatial arrangement. GASP [63] and GAMMA [74] use such
an approach by the genetic algorithm (GA) techniques.
5.1.3 Feature Extraction and Representation
After conformational search, the molecules are subdivided into a set of features,
each feature having the capability to form a particular type of non-covalent interaction with the receptor. There are three main levels of resolution for defining the
features; (i) it may be atom based as implemented in MPHIL [75], GAMMA [74]
and RAPID [69], where 3D atomic position related to the atom type is used as a
feature; (ii) it can be atoms grouped into topological features such as a C = O group
or a phenyl ring; or (iii) it may be function based, where the atoms are assembled
into functional features describing the type of non-bonded interactions with the
receptor. These features are HB acceptor (A), HB donor (D), base (+ve charge pH
7) (P), acid (−ve charge, pH 7) (N), aromatic moieties (rings) (R) and hydrophobic
group (H). We have already discussed these features in Sect. 3 of this chapter. The
third type of feature extraction method is immensely popular and is being used in
many programmes like catalyst [43], Phase [46], HypoGen and HipHop [63].
Different topological features having the same chemical function can fall under
same functional feature category. At the same time, the functional features are
not assigned exclusively for any functional group. For instance, a –OH oxygen can
act as both HB acceptor, a donor and at times may act as negatively charged feature.
Commonly, the functional groups like a negatively/positively charged species, HB
donor and acceptor are represented by their centres, which are nothing but the exact
atom positions. Additionally, HB acceptors and donors are often represented by a
vector that enforces a restriction of bond directionality between the feature on the
binding site of the receptor and the complementary ligand feature. The centre of a
hydrophobic site or an aromatic ring is defined as the centroid of the group.
After extracting the features, depiction of the whole molecule’s structure is
obtained by combining the selected features. These representations are generated
mostly as: (i) 3D point set, where a ligand structure is represented as a group of
categorized points in the 3D space, where each point is linked with a feature, (ii) a
labelled graph, where nodes correspond to the features and the edges correspond to
the relations, or (iii) a set of interpoint distances, where the ligand structure is
represented as a collection of feature points, along with their interpoint distances.
The third type of representation is commonly stored as a n  n distance matrix, n
being the number of atoms.
34
C. Choudhury and G. Narahari Sastry
generation. In some other software, conformational search is parallelly performed
along with pattern identification by retaining the conformers that possess certain
features in a particular spatial arrangement. GASP [63] and GAMMA [74] use such
an approach by the genetic algorithm (GA) techniques.
5.1.3 Feature Extraction and Representation
After conformational search, the molecules are subdivided into a set of features,
each feature having the capability to form a particular type of non-covalent interaction with the receptor. There are three main levels of resolution for defining the
features; (i) it may be atom based as implemented in MPHIL [75], GAMMA [74]
and RAPID [69], where 3D atomic position related to the atom type is used as a
feature; (ii) it can be atoms grouped into topological features such as a C = O group
or a phenyl ring; or (iii) it may be function based, where the atoms are assembled
into functional features describing the type of non-bonded interactions with the
receptor. These features are HB acceptor (A), HB donor (D), base (+ve charge pH
7) (P), acid (−ve charge, pH 7) (N), aromatic moieties (rings) (R) and hydrophobic
group (H). We have already discussed these features in Sect. 3 of this chapter. The
third type of feature extraction method is immensely popular and is being used in
many programmes like catalyst [43], Phase [46], HypoGen and HipHop [63].
Different topological features having the same chemical function can fall under
same functional feature category. At the same time, the functional features are
not assigned exclusively for any functional group. For instance, a –OH oxygen can
act as both HB acceptor, a donor and at times may act as negatively charged feature.
Commonly, the functional groups like a negatively/positively charged species, HB
donor and acceptor are represented by their centres, which are nothing but the exact
atom positions. Additionally, HB acceptors and donors are often represented by a
vector that enforces a restriction of bond directionality between the feature on the
binding site of the receptor and the complementary ligand feature. The centre of a
hydrophobic site or an aromatic ring is defined as the centroid of the group.
After extracting the features, depiction of the whole molecule’s structure is
obtained by combining the selected features. These representations are generated
mostly as: (i) 3D point set, where a ligand structure is represented as a group of
categorized points in the 3D space, where each point is linked with a feature, (ii) a
labelled graph, where nodes correspond to the features and the edges correspond to
the relations, or (iii) a set of interpoint distances, where the ligand structure is
represented as a collection of feature points, along with their interpoint distances.
The third type of representation is commonly stored as a n  n distance matrix, n
being the number of atoms.
34
C. Choudhury and G. Narahari Sastry