Biomedical Engineering Reference
In-Depth Information
this case, primary node colors indicate a specific mechanism of action (e.g., agonists,
partial agonists, inverse agonists, or antagonists) and different potency values are
reflected by color transparency levels. This NSG variant makes it possible to select
subsets of similar compounds from regions of apparent mechanistic heterogeneity
and study structural modifications that lead to mechanism hopping [24].
Furthermore, the NSG formalism has been extended to derive selectivity land-
scapes for compounds active against pairs of targets [19]. Here, potency ratios (log-
arithmic potency differences) are utilized instead of compound potency values. The
topology of the single-target potency-based NSGs and a dual-target selectivity-based
NSG is conserved because it is determined only by compound similarity relationships.
For the representation of selectivity landscapes, per-compound discontinuity scores
are also calculated on the basis of potency ratios. Hence, potency-and selectivity-
based NSGs can be compared for a given compound data set, and different types of
relationships between single-target activity cliffs and selectivity cliffs can be studied.
A logical extension of selectivity landscapes for target pairs is the design of mul-
titarget activity landscapes for compound sets with activity against three or more
targets. Conceptually, this represents a more difficult problem than transforming
activity landscapes into selectivity landscapes because “vectors” of potency informa-
tion for multiple targets must be accounted for and compared in a consistent manner.
A first multitarget landscape representation has recently been introduced [20]. For
this purpose, an NSG variant was designed in which a ternary potency code was
assigned to each compound node on the basis of a potency binning scheme, that
is, by classifying potency values against a given target as highly (“2”), moderately
(“1”), or weakly (“0”) potent. For node scaling, a compound discontinuity score was
calculated over all targets to account for the introduction of SAR discontinuity in
multitarget space. Furthermore, a modified color code was applied to nodes indicat-
ing selected potency profiles. On the basis of this landscape representation, subsets
of compounds can be prioritized for further study that are characterized by a high
degree of SAR discontinuity in multitarget space and that form multitarget activity
cliffs. Subsequently, an alternative multitarget activity landscape design was reported
where compounds were encoded as arrays of pairwise target potency relationships and
classified in self-organizing maps [25]. Furthermore, extensions of SAS maps have
also been introduced to represent multitarget SARs (see above) [18,21]. For example,
pairwise compound activity similarity over multiple targets has been calculated as
Tanimoto similarity of real-valued potency vectors [21].
16.4.3.3 Bipartite Matching Molecular Series Graph
A general caveat for the
interpretability of activity landscape representations in medicinal chemistry is the
use of calculated (Tanimoto) similarities. This is the case because structural rela-
tionships that are based on calculated whole-molecule similarity values are often not
easy to interpret in chemical terms. Accordingly, a molecular network-based activity
landscape design has recently been introduced in which calculated similarities are
replaced with well-defined and immediately interpretable substructure relationships
between active compounds [26]. This activity landscape is termed a bipartite match-
ing molecular series graph (BMMSG; Figure 16.1). Substructure relationships are
