Chemistry Reference
In-Depth Information
in the older marketed drugs and absent in new ones. This may reflect the problem of the
developing resistance observed against older antibiotics. Another example is the absence
of amino acid scaffolds and side-chains in marketed oral drugs. Likewise, the majority of
amino acid scaffolds are exclusive to injectable drugs.
8.3 Learning from Existing Databases
There is a lot to be learned from existing (drug) compound databases in terms of fragments:
which fragments exist, how frequent they are and how the occurrence of one fragment is
related to the occurrence of another, nonoverlapping fragment.
[
27
]
For instance, one can
find single fragments that occur extremely often (e.g. a phenyl ring) or chemical templates
on which some drug classes are based (e.g. benzodiazepines). Fragments which have low
abundance might indicate barely explored parts of chemical space,
[
27
]
potentially interesting
for designing new compounds. Insight can be obtained into preferences regarding chemistry
and also in differences among databases. In the following sections, we will further expand
on this, discussing analysis and evaluation of such databases (Sections 8.3.1 and 8.3.2) and
applications of the findings thereof (Sections 8.3.3, 8.3.4 and 8.3.5).
8.3.1 Analysis of a Single Database
In an effort to identify the common features present in drugmolecules, Bemis andMurcko
[
28
]
analyzed the structures of 5120 drugs extracted from the Comprehensive Medicinal Chem-
istry (CMC) database.
[
29
]
Two types of representation were used, in order to analyze struc-
tures at different levels of detail. At a more general level, properties of the molecular graphs
were analyzed. Since the same graph may represent multiple molecules of similar shape,
the common structure classes are revealed. For example, benzene, hexane and pyridine are
all represented by the same hexagonal graph. In a more detailed analysis, the authors also
considered atomic properties such as atom type, hybridization and bond order. They defined
four nonoverlapping structural units that form a hierarchical description of the molecule:
ring systems, linkers, frameworks and side-chains, as discussed in the subsection
Molecu-
lar building blocks
in Section 8.2.3. The authors justified their choice of this classification
scheme by highlighting its useful features. For example, most frequent frameworks are eas-
ily identified, which may guide future drug design. Moreover, ring systems and linkers can
serve as input for combinatorial library generation. In addition, the simple building blocks
in existing drugs are already useful for checking the overlap between compound libraries.
The graph theoretical approach, as outlined in Section 8.2 and in Figure 8.2, identified a
set of 1179 different frameworks, of which the six-membered ring was the most common
one found. Of all these frameworks, 783 (66%) were unique, i.e. they occurred only once
in the database. However, a small set of only 32 frameworks accounted for 50% of the
drug molecules in the database. Analysis that also considered atomic properties logically
resulted in a more diverse set of frameworks. There were 2506 different frameworks, of
which 1908 (76%) were unique. Not surprisingly, a small set of 41 frameworks accounted
for 1235 drug molecules (24%) in the database. Benzene was the most common framework
found (8.5%). When we think of molecules as a common framework decorated with side-
chains, phenyl and other small rings may be considered side-chains just as well, as in
