A molecular sensor learns to spot ever-changing synthetic cannabinoids

Discerning subtle structural differences among highly similar small molecules remains a major challenge for synthetic receptors. Natural olfactory systems solve this problem using cross-reactive receptor families that combine a conserved docking core with hypervariable specificity domains. Despite the modular nature of nucleic acid aptamers, the variable supporting domains have largely been viewed as inert scaffolds, whereas the conserved regions have been the main focus of functional studies. Here, we demonstrate that these supporting domains actively govern fine-structural selectivity, thereby decoupling specificity from core binding affinity and enabling modular reprogramming of recognition. Guided by this insight, we emulated natural receptor evolution by screening a structure-biased library with randomized peripheral domains, thus building a family of structurally homologous, cross-reactive aptamers against synthetic cannabinoids. When assembled into a machine-learning-guided sensor array, termed an "aptameric nose", this repertoire produces distinct yet correlative response patterns for closely related analogs. The array robustly discriminates these compounds and adaptively identifies emerging new variants. Through modular reprogramming, we engineer artificial receptor families that keep pace with continuously diversifying chemical threats, offering a powerful analytical tool for forensic, clinical, and environmental monitoring.