Dual Molecularly Imprinted Mesoporous Organosilica Nanoprobes for AI-Assisted Glutathione Redox Ratio Sensing

Abstract Molecular recognition nanomaterials that directly decode mixture composition remain underdeveloped because chemically related analytes often produce nonlinear and partially overlapping optical responses. Here, dual molecularly imprinted mesoporous organosilica nanoprobes are developed for AI-assisted determination of the glutathione redox ratio. Zwitterion-embedded mesoporous organosilica nanocavities are independently imprinted with reduced glutathione (GSH) or glutathione disulfide (GSSG), followed by pore confinement of blue-emissive graphene quantum dots as fluorescence transducers. The resulting MIMO_GSH-Q and MIMO_GSSG-Q probes generate complementary raw fluorescence fingerprints from mixed GSH/GSSG samples. A deep neural network trained on paired microplate fluorescence intensities performs inverse prediction of τ = [GSH]/[GSSG]. A single test mixture at τ = 0.60, absent from the training ratios, yields a mean predicted ratio of 0.575. These results provide a proof-of-concept demonstration of two-channel ratio inference under the tested conditions, while broader external validation is required to establish its range of applicability. This work demonstrates a trainable nanoporous recognition platform that transforms selective nanoconfined fluorescence responses into quantitative redox-ratio information, providing a general route toward AI-decodable sensing nanomaterials.