Data-Driven Design of Biofunctional Molecularly Imprinted Polymers for Precision Diagnostics

Conventional imprinted receptors design is laborious and reliant on trial-and-error experimentation. To overcome existing challenges, this study integrates molecular mechanics/molecular dynamics (MM/MD) and machine learning (ML) to design molecularly imprinted polymer nanoparticles (nanoMIPs), aiming to uncover synergies and gaps between physics-informed and data-driven approaches. By systematically comparing and experimentally evaluating MM/MD formulations with ML-predicted recipes, we establish a multimodal framework for cross-validation, method selection, and accelerated development of biomimetic materials for precision cancer diagnostics. A monomer library is screened using hybrid MM/MD method, leading to a generalizable nanoMIP formulation for detecting cancer biomarkers. In parallel, we introduce an ML algorithm that predicts nanoMIP formulations, reducing formulation prediction time for new epitopes from months of conventional computational/experimental screening to seconds after model training. Trained on epitope fragments, the ML framework supports epitope-mediated imprinting by generating recipes tailored to the physicochemical properties of targeted peptides. The curated dataset and integrated MM/MD-ML workflow developed in this study, capture molecular interactions beyond experimental reach, while experimental synthesis and evaluation ensure alignment between formulation predictions and nanoMIP performance under experimentally relevant conditions. This multimodal, feedback-loop-based framework improves predictive accuracy of imprinted polymer formulations, provides mechanistic insights, and supports hypothesis-guided design of advanced nanoMIPs.