Searching for Coordinated Activity Cliffs Using Particle Swarm Optimization

Abstract

Activity cliffs are formed by structurally similar compounds having large potency differences. Coordinated activity cliffs evolve when compounds within groups of structural neighbors form multiple cliffs with different partners, giving rise to local networks of cliffs in a data set. Using particle swarm optimization, a machine learning approach, we systematically searched for coordinated activity cliffs in different compound sets. Regardless of the global SAR characteristics of these data sets, coordinated activity cliffs introducing strong local SAR discontinuity were identified in most cases. Compound subsets forming coordinated activity cliffs represent centers of SAR discontinuity and have high SAR information content. Through particle swarm optimization guided by subset discontinuity scoring, compounds forming the largest coordinated activity cliffs can automatically be extracted from large compound data sets.