Biomedical Engineering Reference
In-Depth Information
the similarity conundrum. In the context of activity landscapes, experimental data
variability refers to potency measurements. A systematic analysis of activity cliff for-
mation using alternative compound potency measurements has recently been carried
out for BindingDB compounds [43,44]. Applying the definition of discrete activity
cliffs stated above, it was found that activity cliff distributions in many different data
sets changed significantly when means of multiple available IC 50 or K i values were
used or minimum and maximum values instead. On average, only about 50% of activ-
ity cliffs were conserved when multiple measurements were utilized in different ways.
In cases where all available potency values fell within an order of magnitude, fewer
activity cliffs were detected than when the data were more variable. Furthermore,
K i measurements yielded consistently fewer activity cliffs than IC 50 measurements
[44]. Different from IC 50 , K i values represent equilibrium constants and are thus not
dependent on assay conditions and are less error-prone. Taken together, these findings
show that approximate potency measurements and increasing data variability gen-
erally produce larger numbers of activity cliffs than more accurate potency values.
Thus, in addition to the choice of molecular representations, the types of experimental
potency values and their confidence level must be considered carefully in the analysis
of activity cliffs.
16.5.4 Different Types of Activity Cliffs
In addition to “standard” activity cliffs formed by pairs of similar compounds, other
types and variants of activity cliffs have been introduced. For example, a special
category of activity cliffs are “R-cliffs” in analog series that result from R-group
replacements at specific substitution sites [33]. R-cliffs have been defined on the
basis of matrix extensions of R-group tables (see above). Furthermore, going beyond
conventional single-target activity cliffs, selectivity cliffs [19] and multitarget activity
cliffs [20] have been introduced. Selectivity cliffs are formed by pairs of compounds
with significantly different potencies against one or both targets of a target pair and
are the cardinal feature of selectivity landscapes. In addition, multitarget cliffs are
formed by pairs of compounds with different potencies against series of targets. In
Figure 16.3, the formation of single- and multitarget activity cliffs and selectivity
cliffs is illustrated. Multitarget cliffs are “directed” if one compound has consistently
high potency against its targets, or “undirected” if the cliff partners display differential
(inverse) potency against one or more targets [20].
Considering the molecular representation dependence, activity cliffs have been
specified that are consistently formed in different chemical reference spaces called
consensus activity cliffs [45]. Such consensus cliffs are hence thought to be the least
descriptor-dependent. Another type of consensus activity cliff might be conceptu-
alized considering the influence of experimental measurements and data variability.
These experimental consensus cliffs would be conserved when alternative potency
measurements and multiple potency values are utilized. It should be interesting
to compare representation-independent activity cliffs with experimentally invariant
cliffs and explore potential overlap between these populations in different compound
data sets.
Search WWH ::




Custom Search