Biomedical Engineering Reference
In-Depth Information
database and four combinatorial libraries were compared [105]. Physicochemical
properties included the six medicinally relevant properties discussed above and
molecular fingerprints of different design [125,135]: namely, MACCS keys, TGD, and
graph-based three-point pharmacophores (GpiDAPH3) [129]. The latter fingerprints
are graph-based three-point pharmacophores employing any set of three possible
atom types: pi system, donor, or acceptor. The use of multiple criteria allowed the
authors to obtain a more comprehensive analysis of the density, coverage of chemical
space scaffold, content, diversity, and structural similarity of combinatorial libraries
and natural products compared to other collections, particularly known drugs. The use
of different representations covering physicochemical properties, structural finger-
prints, and molecular scaffolds has been used to compare other compound databases,
such as natural products from TCM and DOS libraries [48,72].
10.5 RECENT TRENDS IN COMPUTATIONAL APPROACHES TO
CHARACTERIZE COMPOUND LIBRARIES
As discussed in the preceding sections, the concept of the chemical space is widely
employed in drug discovery and other research disciplines. However, there is not a
unique definition of this concept. Similarly, the chemoinformatic characterization of
compound libraries is a common practice in several research groups. Nevertheless,
similar to the concept of chemical space, there is no unique, “best” protocol to
use to conduct the analysis. Chemoinformatic characterization and mining of the
chemical space is not a solved problem. The high dependence of chemical space with
structure representation and the potential to have very high dimensional chemical
spaces represent major challenges to mine, visualize, and extract useful information.
Therefore, the research community is constantly developing and applying novel ways
to address the major issues of characterization of compound libraries.
At the time of writing (April 2012), one of the latest advances in chemoinformatic
approaches to mining and visualizing the chemical spaces of compound databases
is the generation of delimited reference chemical spaces (DRCSs) [39,136]. In this
work, Le Guilloux et al. analyzed 6.6 million unique screening compounds collected
from 73 chemical providers and generated a representative contour that represents
the overall shape of the chemical space of a very large proportion of the molecules
analyzed in this work. Therefore, the DRCS method delimits the densest subspace
spanned by a reference library in a reduced two-dimensional continuous space and
has been used as a basis to develop subspace-specific diversity indices [136].
A recent development in the depiction of chemical space in low dimensions is
represented by the latent trait model. This method was employed to reduce the dimen-
sionality of a fingerprint array from 166-dimensional space to two-dimensional space,
where it could be plotted and visualized. The chemical space of five combinatorial
libraries and other reference collections was visualized using this method [70].
To generate visual representations of the chemical space using simple, yet mean-
ingful descriptors, the research group of Reymond proposed a set of 42 integer
value descriptors of molecular structure called molecular quantum numbers (MQNs)
