Biomedical Engineering Reference
In-Depth Information
obtained through calculations of Tanimoto similarity of real-valued vectors contain-
ing compound potency values against three targets. In analogy, for the computation
of SALI scores, this multitarget activity similarity replaces single-target potency
differences for each compound pair.
16.4.3 Molecular Networks
Recent activity landscape design efforts have also focused on molecular network
representations. In general, in these types of landscapes, nodes represent compounds
and edges pairwise compound similarity relationships or, alternatively, activity cliffs.
16.4.3.1 SALI Graph
The SALI graph is based on the SALI formalism and is
an activity cliff-centric representation of an activity landscape (Figure 16.1) [11].
Here, edges indicate activity cliffs and are drawn as an arrow from the less to the
more potent compound of a pair. Two compounds are connected by a directed edge
if their SALI score exceeds a given threshold value, for example, a score greater than
60, 70, or 80% of all pairwise scores in a data set. By applying different threshold
values, networks of activity cliffs of increasing magnitude are obtained. The SALI
formalism can also be applied for the assessment of computational SAR models [22].
For this purpose, it is determined how many edges in a SALI graph are predicted
correctly by a given computational model. A SALI curve is calculated that reports
the percentage of correctly directed edges as a function of the SALI threshold values
applied to generate the SALI graphs.
16.4.3.2 Network-like SimilarityGraph
The network-like similarity graph (NSG)
design represents a prototype of a similarity-based compound network as an activity
landscape (Figure 16.1) [10]. Themajor characteristic of this landscape representation
is its ability to reveal relationships between global and local SAR features in large data
sets. In this network, nodes represent compounds that are color-coded according to
their potency values and scaled in size according to the per-compound discontinuity
score (as discussed above) that quantifies the individual contribution of each molecule
to local SAR discontinuity. In addition, edges represent pairwise Tanimoto similarity
relationships. An edge is drawn between two nodes if their similarity value exceeds
a predefined threshold. The NSG can be annotated with additional information, for
example, results of cluster analysis and compound cluster discontinuity scores to
characterize compound subset SARs [10]. In an NSG, combinations of large red
(highest potency in a data set) and green (lowest potency) nodes connected by edges
mark the most significant activity cliffs in the set. In addition, NSGs identify different
local continuous or discontinuous SAR environments. Their combination provides
a view of the global SAR character of a compound data set, which can also be
compared with global SARI scoring. NSG analysis is also applicable to SAR mining
of high-throughput screening (HTS) data [23].
Several extensions of the NSG concept have been introduced. For example, com-
pound potency measurements have been combined with molecular mechanism-of-
action information for receptor ligands by modifying the node color code [24]. In
