Biomedical Engineering Reference
In-Depth Information
or combinations that are SAR hotspots. Furthermore, given their hierarchical orga-
nization, CAGs also identify SAR holes (different from those in three-dimensional
landscapes, as discussed above), that is, substitution site combinations that have not
yet been explored [34]. Moreover, side-by-side comparison of CAGs for series of
analogs active against several targets enables the study of multitarget SARs [35].
16.5 DEFINING AND IDENTIFYING ACTIVITY CLIFFS
Having reviewed different approaches to activity landscape design, the following
discussion focuses on activity cliffs, the most prominent features of landscape rep-
resentations. In the literature, the term
activity cliff
probably first appeared in 1991
[36]. If we consider the general definition of an activity cliff (see above), it is not sur-
prising that the activity cliff concept has become rather popular in chemoinformatics
and medicinal chemistry given its chemically intuitive nature and immediate link to
SAR information.
16.5.1 Similarity and Potency Criteria
However, the definition of activity cliffs also raises two important questions. First,
how similar must compounds be to qualify as cliff partners? Figure 16.2 illustrates
potential ambiguities of activity cliffs designated on the basis of calculated similarity
values. Two pairs of compounds forming activity cliffs are shown. In both instances,
Tanimoto similarity is high, but there is a clear difference between these compound
pairs. One cliff is formed between two very closely related analogs, whereas the
compounds forming the other cliff display a large chemical change in their core
structures. Thus, from a chemistry perspective, the structural/similarity relationships
characterizing these cliffs are certainly not equivalent. Second, what are acceptable
potency value ranges and significant potency differences? A cliff formed between
two compounds with 10 nM and 10
M potency would not be considered comparable
to a cliff formed by two compounds with 1
M potency, although in
both cases there is a 100-fold difference in potency and the potency value ranges
overlap. From an SAR point of view, the first cliff would be more interesting than
the other because it involves a highly potent compound rather than two weakly
potent compounds.
These examples illustrate potential complications associated with the activity cliff
concept. There currently are no generally accepted similarity and potency criteria
to define activity cliffs. However, for any activity cliff analysis, such criteria must
be clearly specified (which is often not the case in the literature). Of course, the
dependence of activity landscape topology on chosen molecular representations and
similarity measures also applies to the formation of activity cliffs, and their distri-
bution is expected to change when alternative molecular representations are used.
This has been well illustrated for activity landscape models of a given data set calcu-
lated with different fingerprints or other descriptors [14,16]. Hence, if comparisons
between different data sets are made, it is essential to use the same representation
M and 100
