Biomedical Engineering Reference
In-Depth Information
case because a prototypic activity landscape is formed by compounds sharing the
same specific activity but having different potencies. Hence, chemical reference
space is primarily not required to account for bioactivity-specific features, but must
reflect different degrees of molecular similarity (structural relatedness). However
designed, a chemical space representation forms a foundation of activity landscape
generation, as distances between compounds in chemical space must reflect pairwise
similarity relationships in a meaningful manner. Chemical space representations of
different complexity might be chosen. For example, if we consider the calculation of
fingerprint distances, which are utilized, for example, in the context of the SAS map,
SARI, and SALI, “coordinate-free” reference spaces of relatively low complexity
are obtained that are defined solely by all pairwise compound distances in a given
data set. In principle, any chemical reference space is transformed into an activity
landscape by adding a biological “hypersurface” to it that accounts for compound
potency information. Depending on the nature of the landscape design, this can
be accomplished in different ways: for example, by calculating activity similarity
from potency values. However, the well-known variance of many chemical space
representations represents the most critical variable in activity landscape design.
16.3.2 Similarity Assessment
Activity landscape methods typically rely on calculated whole-molecule similar-
ity, which is greatly influenced by chosen molecular (chemical space) representa-
tions. When different descriptors such as binary fingerprints or numerical molecular
property descriptors are used, pairwise compound similarity or dissimilarity (dis-
tance) relationships usually change, often in a significant manner. Because activity
landscape modeling involves exhaustive enumeration of pairwise compound sim-
ilarity/dissimilarity relationships, even moderate changes in pairwise relationships
might distort landscapes. Ensuing changes in activity landscape topology inevitably
alter the SAR information content of alternative landscape representations of a given
compound data set. For example, the higher the chemical “resolution” of chosen
descriptors, the more dissimilar compounds will on average appear to be, which
influences landscape topology in a systematic manner. The representation depen-
dence of similarity/dissimilarity value distributions must be taken into account when
analyzing and comparing activity landscape models (see below). The choice of molec-
ular representations typically has a greater influence on the information content and
interpretability of activity landscape models than alternative similarity metrics. As is
the case for chemical reference spaces in chemoinformatics, there are no generally
preferred solutions. However, because in medicinal chemistry SAR information is
ultimately evaluated on the basis of two-dimensional molecular graphs, graph-based
measures of structural similarity such as fragment or topological fingerprints [6] often
yield robust landscape views that are interpretable from a chemical perspective [1].
For any chosen molecular descriptors, it is essential for activity landscape analysis
that calculated similarity values are chemically intuitive and readily interpretable, as
discussed in the following.
