Combating Counterfeit Drugs via Machine Learning-Enabled Array Screening of Multilayer Evolutionary Combinatorial Libraries

Counterfeit drugs are a global issue that has a serious impact on patient morbidity and mortality. Driven by nonspecific cross-reactivity, sensor arrays enable the concurrent discrimination of structurally related drug molecules. Nevertheless, rapidly generating sensor element libraries without a labor-intensive synthesis remains a major challenge. Herein, we present a machine learning-guided, three-layer screening strategy to identify the minimal optimal combination of sensing elements to combat counterfeit nonsteroidal anti-inflammatory drugs (NSAIDs), using a combinatorially designed library with 100 candidates. Following screening, a pruned 5-element array was successfully constructed, achieving 100% accuracy in distinguishing among nine NSAIDs and their analogs. Furthermore, the pruned arrays successfully achieved quantitative and multiplexed differentiation of two key NSAIDs. Notably, this strategy accurately discriminated five commercially available over-the-counter (OTC) NSAIDs from two counterfeit counterparts, achieving 100% accuracy within 5 min. These findings pave the way for constructing combinatorial sensing libraries for array-based screening and establish a foundation for a wide range of drug authenticity verification.