activity cliff
Imagine walking across a gently sloping field and suddenly stepping off a cliff. Most small structural tweaks to a molecule nudge its activity a little; an activity cliff is the rare spot where a tiny change sends potency plunging — or soaring — by orders of magnitude.
Technically, an activity cliff is a pair (or small group) of very similar compounds, a matched molecular pair, whose biological activities differ far more than their structural similarity would predict. The cliff reveals that something specific and sensitive is happening at that position in the binding site: the change might create or destroy a key hydrogen bond, clash sterically with a wall of the pocket, or force the ligand into a different conformation.
Activity cliffs are gold for the chemist because they pinpoint exactly where the molecule is talking to the target most intimately, and they are the natural enemy of smooth predictive models, which assume similar structures have similar activity. The caveat is that an apparent cliff can also be an artifact — an assay error, the wrong tautomer or stereochemistry, or an undetected change in binding mode — so a surprising cliff deserves confirmation before it is built upon.
Replacing an amide NH with N-methyl abolishes a hinge hydrogen bond and drops potency a thousand-fold — a textbook activity cliff that tells the chemist that NH is essential.
A near-identical structure with wildly different activity flags a critical interaction.
Activity cliffs are exactly the cases QSAR and machine-learning models struggle with, because those methods lean on the assumption that nearby structures have nearby activities.