Biomedical Engineering Reference
In-Depth Information
methodology is that chemical similarity of active compounds is replaced by ligand-
target interaction similarity. In other words, although Tanimoto similarity is also
calculated for ISAC-based SALI scoring, a unique molecular representation is uti-
lized: a ligand-protein interaction space, which is distinct from conventional chemical
reference spaces.
16.6 ACTIVITY CLIFF SURVEY
One of the most interesting questions concerning activity cliffs includes how often
they actually occur in sets of bioactive compound and how they might be distributed
over ligands of different target families. For assessing the relevance of activity cliff
analysis inmedicinal chemistry, these are important points to consider. In addition, it is
attractive to investigate whether there might be molecular building blocks (structural
motifs) that would be recurrent in activity cliffs.
16.6.1 Frequency of Cliff Formation
To determine how frequently activity cliffs occur, an analysis of BindingDB [43] and
ChEMBL [48] compound data sets has recently been carried out (adhering to the
definition of discrete activity cliffs referred to above) [40]. Based on this analysis,
about 12% of all active compounds were found to be involved in the formation of at
least one or two activity cliffs of considerablemagnitude. Only about 4%of all activity
cliffs were multitarget cliffs, and almost all of them were directed. Thus, although
activity cliffs are formed with notable frequency, the majority of currently available
cliffs are single-target cliffs, and cliff-forming compounds with different selectivity
for multiple targets are extremely rare. In this study it was also determined that cliffs
were distributed similarly over different target families, perhaps unexpectedly so.
However, this observation suggests that activity cliffs will with high likelihood be
identified in compound data sets with diverse bioactivities.
16.6.2 From Isolated to Coordinated Activity Cliffs
When analyzing activity cliffs systematically, one can often observe that they are not
formed in isolation (i.e., by pairs of compound without structural neighbors) but that
there are clusters of cliffs involving multiple compounds. To put these observations
on a quantitative basis, a hypothetical data structure termed an activity ridge was
introduced as a query for data mining [39]. The ridge structure was proposed to
consist of a set of five or more compounds having nanomolar potency within an order
of magnitude and another set of at least five compounds with 100-fold higher or lower
potency, also within an order of magnitude. The compounds in these two sets would
then consistently form pairwise (“combinatorial”) activity cliffs. In a systematic
search of 242 different compound data sets, a total of 125 activity ridges were indeed
identified in 71 sets [39]. These ridge structures contained between 10 (minimally
required) and 70 cliff-forming compounds. For comparison, in addition to 125 activity
Search WWH ::




Custom Search