From empirical imprinting to programmable synthetic receptors: a perspective on data science and machine learning for...

Molecularly imprinted polymers (MIPs) are porous materials generated by templated polymerisation, in which functional and crosslinking monomers are organised around a target molecule and fixed within a polymer network to create high-affinity, selective recognition sites for the chosen analyte. The appeal of MIPs lies in their robustness, low-cost and chemically versatile nature, in addition to directly encoding selective molecular recognition into polymer networks while retaining the thermal, mechanical, and processing advantages of synthetic materials. Yet, despite being discovered a century ago, MIP development remains dominated by empirical, target-by-target optimisation. The field therefore presents a compelling challenge for data science and machine learning: MIP performance emerges from a high-dimensional coupling of material, polymer synthesis and processing parameters but this complex design space is still only sparsely sampled. In this Perspective, we argue that the next step for MIPs is not simply better prediction of pre-polymerisation interactions, which is traditionally used to predict optimal composition. Pre-polymerisation metrics are weak predictors of functional performance once MIPs are synthesized and applied; therefore, a broader shift towards data-driven polymer design which considers manufacturing constraints from the outset is needed. We discuss our future vision for the field and how structured datasets, high-throughput screening, computational modelling, Bayesian optimisation and interpretable machine learning enable the move from empirical recipes towards programmable synthetic receptors. Sensing and sustainable manufacturing routes will be covered, placing particular emphasis routes on exploiting the robustness of MIPs to facilitate high-throughput production and biocompatibility challenges.