Biomedical Engineering Reference
In-Depth Information
natural products from a commercial vendor, diverse compounds from the academic
synthetic chemistry community, and general screening compounds froma commercial
source. Data sets were obtained from the work of Clemons et al. [85]. In all data sets,
the most frequent scaffold is benzene, as reported previously for other collections
of drugs and several other data sets [72,94,105,116]. The flavone and coumarine
scaffolds that are frequent in this natural product collection (Figure 10.5) are also
found frequently in the databases of natural products implemented in ZINC [72,105].
The structural complexity of some of the frequent scaffolds in the synthetic collection
obtained from academic groups is remarkable.
Similar to the dependence of the chemical space with structure representation
using descriptors, scaffold analysis depends on the definition of scaffold. It has
been pointed out that a preferred scaffold representation is objective, invariant, and
not dependent on the data set. There are various ways to derive the scaffold of
a molecule computationally in a systematic and consistent manner that have been
reviewed elsewhere [92,99]. One group of commonly used representations are the
atomic frameworks
of Bemis and Murcko, defined as the union of ring systems and
linkers in a molecule [94]. This definition is similar to the
topological scaffolds
of
Xu, defined as ring bonds and linker bonds but not chain bonds [96,117], or the
cyclic systems
of Xu and Johnson [118,119], which result by iteratively removing
the side chains of the molecule (Figure 10.6). The cyclic systems are part of the
chemotype methodology developed by Johnson and Xu and are calculated using the
program Molecular Equivalent Indices (MEQI). The latter approach has been used
successfully to classify collections of combinatorial libraries, drugs, natural products,
and other compound databases [72,105,116,120].
The Johnson-Xu approach decomposes compounds in terms of characteristic
structural patterns of variable resolution and complexity called
chemotypes
and pro-
vides tools for a hierarchical classification based on the chemotypes [18,119]. Fig-
ure 10.6 shows that a cyclic system skeleton is obtained if all atoms are set to a single
atom type (e.g., carbon) and all bonds are set to single bonds in the cyclic system.
Deleting all atoms attached to two other nonhydrogen atoms from the cyclic system
skeleton leads to a reduced cyclic system skeleton, which is a low level of structural
resolution. In Figure 10.6 the curved lines designate multibond links connecting the
ellipsoidal objects, which represent general ring structures, while the corresponding
straight lines represent single bonds. Note the hierarchical relationship at different
levels of structural resolution.
Figure 10.7 shows a simple example of the hierarchical relationship among chemo-
types at four different levels of resolution [111]. This figure depicts the classification
of five structures obtained from in-house combinatorial libraries and deposited in
PubChem. The five structures at the bottom of the figure share the same reduced
cyclic system skeleton chemotype, which corresponds to the root node of the hierar-
chical tree. Moving from the root node to the leaf nodes corresponds to increasing
structural resolution. At the level of the hierarchy lying just below the root node, the
structures are classified into two cyclic system skeleton chemotypes. Three chemo-
types are shown at the next lower level, which corresponds to cyclic systems. The
lowest level of the hierarchy, which contains the leaf nodes, corresponds to the highest
