Generic strategy for high-performance bio-based imprinted membranes free of non-specific adsorption: machine learning-assisted sensing.

A rapid, green, and cost-effective strategy is proposed for the preparation of bio-based molecularly imprinted membranes (MIMs) by studying diverse functional biopolymers, including chitosan, sodium alginate, carboxymethyl cellulose, sulfonated cellulose nanocrystalline, cellulose acetate, and gelatin. These MIMs were tested for tetracycline and caffeic acid (Caf) templates. The MIMs were prepared in less than 2 h, without the need for complex synthesis, toxic, or expensive reagents. Furthermore, various approaches were introduced to eliminate the non-specific adsorption (NSA) for the first time, using covalent and non-covalent crosslinking. After appropriate selection of the biopolymer and crosslinker, the affinity of the non-imprinted membranes, free of cavities, toward the templates became negligible, thereby confirming the improvement in imprinting performance and the suppression of NSA. Coupling these NSA-free MIMs with smartphone colorimetric detection offers rapid, cost-effective, and on-site sensing. Sensor arrays were developed for the detection of Caf in pears, plums, and apples. The RGB spectral data was processed using machine learning. The artificial neural network model showed excellent regression performance, with high R2 (0.989–0.970) and low RMSE (0.01–0.05), confirming the strategy’s precision and the MIMs selectivity efficiency in complex matrices. This work provides a pathway to transition from conventional synthetic to sustainable artificial antibodies for an ultra-selective, green, cost-effective, and efficient future for imprinting technologies.