Activity cliffs are defined as pairs of similar compounds with substantial differences in activity (Molecular Diversity, 19, 1021–1035 (2015)). Are there computational tools for the identification of activity cliffs between two analog compounds? This paper lists chemotypes that have a high propensity to induce activity cliff.
A 2015 paper  looked at the ability to predict activity cliffs structurally using docking models that could combine experimental and computational information about the conformation of the receptor and/or binding molecule. They found that they could achieve reasonable prediction accuracy and planned to implement further improvements to make a more robust predictor model.