Biomedical Engineering Reference
In-Depth Information
than 5000 types of descriptors are available [57]. The choice of descriptors used to
analyze compound data sets gives rise to different types of chemical spaces. Varnek
and Baskin have pointed out that “unlike real physical space, a chemical space
is not unique: each ensemble of graphs and descriptors defines its own chemical
space” [37].
Sections 10.4.1 to 10.4.4 focus on a discussion of representative examples of the
chemoinformatic characterization of the chemical space and molecular diversity of
compound libraries using different structure representations.
10.4.1 Physicochemical Properties and Medicinally Relevant
Chemical Spaces
Physicochemical properties have been used to develop the classical rules to define
drug-like [58] and lead-like criteria [59] and Congreve's rule of 3 for fragment-based
lead discovery [60]. It is worth noting that the seminal Lipinski's rule of 5 [58] has
been revised over the past decade. For example, Walters et al. present a comprehen-
sive review of studies showing the changes over the years in the physicochemical
properties of compounds synthesized for drug discovery programs [61]. It is largely
documented that “new molecular entities are moving away from the traditional drug
space” [62,63] and that “as new targets emerge and optimization tools advance, the
oral drug like space might expand” [64]. Dow et al. noted that the current “biologi-
cally relevant chemical space” can be considerable skewed by the uneven exploration
of chemical space by chemical synthesis [35].
Physicochemical properties frequently used to describe chemical libraries include
molecular weight (MW), number of rotatable bonds (RBs), hydrogen-bond accep-
tors (HBAs), hydrogen-bond donors (HBDs), topological polar surface area (TPSA),
and the octanol/water partition coefficient (
S
log
P
), which are properties commonly
used as descriptors to represent lead-like, drug-like, or medicinally relevant chemical
spaces. An advantage of using these properties to compare databases is that they
are intuitive and straightforward to interpret. Characterization of compound libraries
using additional properties related to ADME attributes are being explored increas-
ingly (see Figure 10.1). This is exemplified by the work of Wager et al., who analyzed
the ADME space of marketed central nervous system (CNS) drugs and an in-house
collection [65] and the rich review of Ritchie et al. covering tools to visualize and
represent ADME-related properties of data sets [51].
A number of comparisons of chemical data sets in the medicinally relevant chem-
ical spaces have been published for different purposes. For example, the chemical
space of 15 combinatorial libraries organized in three bis-diazacyclic “libraries from
libraries” (LOLs) was recently compared to approved drugs using PCA and the distri-
bution of six medicinally relevant properties: MW, RB, HBA, HBD, TPSA, and
S
log
P
(see above). The LOL strategy is a solid-phase DOS technique [66], where multi-
ple scaffolds are generated from the same starting material (Figure 10.2a). Increasing
skeletal diversity is known to be a very efficient way to increase structural diversity
[67]. LOLs are designed to have different physicochemical and biological properties
from those of the parent library [68]. This approach has been employed to synthesize
