Frequency of occurrence and potency range distribution of activity cliffs in bioactive compounds

We have systematically identified activity cliffs in bioactive compounds for which high-confidence potency data were available. Different molecular representations were utilized and similarity and potency difference criteria for activity cliffs were clearly defined. Cliff formation was studied across the global potency range observed for qualifying bioactive compounds. Depending on the specific representation of activity cliffs, between ∼22% and 34% of all active compounds meeting the data selection criteria were involved in the formation of at least one activity cliff spanning a potency difference of at least two orders of magnitude. However, these cliffs involved only between ∼4.7% and ∼5.7% of all compound pairs meeting the similarity criteria. Hence, in light of these findings, the formation of well-defined activity cliffs is a relatively rare event. Moreover, the potency range distribution of activity cliffs was analyzed in detail, revealing that most activity cliffs were formed between compounds with micro- and nanomolar potency, consistent with the global distribution of potency data. On the basis of our analysis, we propose a general definition of activity cliffs for data mining.