Biomedical Engineering Reference
In-Depth Information
three-dimensional models also reveal characteristic topologies that correspond to
specific SAR features of compound data sets, in analogy to idealized models. Three-
dimensional models are designed primarily to provide global views of SAR features.
In addition, these three-dimensional models are particularly useful as a diagnostic
tool to illustrate the in-part dramatic dependence of landscape topology on chosen
molecular representations [14]. Of course, three-dimensional landscapes might also
be rationalized as nonlinear QSAR-type model and hence could be considered for
activity prediction. However, in our experience, interpolation of larger surfaces from
sparsely distributed potency data is in general not sufficiently accurate to predict
compound potencies, in particular, in activity cliff regions.
16.4.2 SAS Maps
SAS maps, first reported by Shanmugasundaram and Maggiora in 2001 [8], are
a prototype of graph-based activity landscape models. In SAS maps, systemati-
cally computed pairwise structural and activity similarity (see above) are plotted
against each other such that each data point reports a pairwise compound comparison
(Figure 16.1). These data points are then color-coded according to the potency value
of the more active compound of each pair. This data representation can be modi-
fied in different ways. For example, compound potency differences or the sum of
compound potency values might be utilized instead of calculated activity similarity
[9]. A SAS map is best rationalized to consist of four sections that capture different
SAR characteristics (Figure 16.1). The upper-left section contains compound pairs
with high activity similarity and low structural similarity and hence corresponds to
a scaffold hopping region. The upper-right section contains compounds with high
structural and high activity similarity such as analogs with comparable potency. The
lower-left section is characterized by the presence of low structural and low activity
similarity and is thus not very informative from an SAR perspective. By contrast,
compound pairs falling into the lower-right section have high structural similarity and
low activity similarity. Accordingly, this section represents the activity cliff region of
an SAS map.
Yongye et al. have introduced a number of modifications and extensions of the
original SAS map. For example, consensus SAS maps have been generated by cal-
culating mean similarity values for different two- and three-dimensional fingerprints
[16] in order to balance the dependence of landscape topology on individual molecular
representations. Another variant utilizes logarithmic potency differences against two
or three targets for compound pairs instead of single-target activity similarity, thereby
producing two-dimensional dual activity difference (DAD) [17] or three-dimensional
triple activity difference (TAD) maps [18]. In these cases, dots representing compound
pairs are color-coded by fingerprint similarity. DADmaps follow the ideas underlying
the design of selectivity landscapes [19] and TAD maps the principle of multitarget
activity landscapes [20], another recent extension of the activity landscape concept
(see below), which is also addressed through the introduction of structure multiple
activity similarity (SmAS) maps and the structure multiple activity landscape index
(SmALI), a variant of the SALI formalism [21]. In this case, activity similarity is
