Biomedical Engineering Reference
In-Depth Information
16.4 ACTIVITY LANDSCAPE REPRESENTATIONS
In the following, activity landscape models of different design are introduced. Cur-
rently available activity landscapes essentially can be divided into “normal” graphs,
molecular networks, and three-dimensional models. Different types of activity land-
scapes are displayed in Figure 16.1.
16.4.1 Three-Dimensional Models
Probably the most intuitive way to rationalize an activity landscape is to envision the
addition of the biological hypersurface as a third dimension to a two-dimensional
projection of chemical space. Then, activity landscape views become rather similar to
geographical landscapes, and their topology reminds us of plains, mountains, valleys,
and so on. The attractiveness of such idealized (theoretical) three-dimensional land-
scapes for the discussion and rationalization of SAR features had been recognized
long before comparable modes were derived for actual data sets [12]. In an ideal-
ized three-dimensional model, hypothetical potency information is represented as a
contiguous surface without indicating the positions of individual compounds (which
are located in the two-dimensional projection of chemical space). In such three-
dimensional models, gently sloped areas represent regions of local SAR continuity
where gradual changes in chemical structure are accompanied by small or moderate
changes in potency. In regions of SAR continuity, scaffold hopping [13] can occur;
that is, a transition can be facilitated from one chemotype to another without a dra-
matic change in potency. It should be noted that for compound subsets forming regions
of local SAR continuity, it might be possible to build linear models of activity as a
function of structural modifications, consistent with the classical QSAR paradigm.
By contrast, rugged areas represent regions of local SAR discontinuity where small
changes in chemical structure lead to large changes in potency. In these regions, large
peaks represent activity cliffs, the extreme form of SAR discontinuity. In Figure 16.1,
an idealized “variable” activity landscape is shown that combines smooth and rugged
regions and hence corresponds to global SAR heterogeneity.
Equivalent three-dimensional models can also be built on the basis of “real data,”
also shown in Figure 16.1. Despite the intuitive nature and attractiveness of theoretical
three-dimensional landscape views, the first real data models have only recently been
introduced [14]. For various compound data sets, coordinate-free chemical reference
spaces formed by pairwise fingerprint distances were generated and projected onto
the
x
-
y
plane of a coordinate system through multidimensional scaling [15]. From
sparsely distributed potency values of test compounds, a contiguous surface was
generated along the
z
-axis through the application of interpolation functions. The
potency surface is color-coded according to surface elevation, applying a continuous
color spectrum (from green to red). Thus, prominent activity cliffs appear as red peaks
(Figure 16.1). White regions correspond to interpolated surface area where compound
data are lacking. These regions can be regarded as “SAR holes” for a given data
set. Similar to idealized three-dimensional landscape views, compound data-based
