Active Learning for the Discovery of Antiviral Polymers

ABSTRACT The development of nanomedicines has long relied on scientific intuition within finite chemical space, but machine learning offers a data‐driven approach to explore far broader chemical landscapes. Advances in high‐throughput polymer synthesis and screening now enable the generation of datasets large enough to train predictive models to accurately link polymer structure with biological function. The next challenge is using machine learning to guide iterative development of nanomedicines through active learning. Here we demonstrate an active learning workflow for the design of antiviral polymers. Using molecular descriptors derived from polymer composition and monomer structure, a machine learning model was trained on an experimental dataset of antiviral polymer activity. The trained model was coupled with unsupervised clustering of monomers to explore chemical diversity and predict antiviral activity across a virtual library of up to 500,000 new polymers. By calculating the expected improvement function, active learning identifies optimal candidates for synthesis to efficiently explore chemical space and optimize antiviral activity. This communication highlights how machine learning and active learning can serve as practical, accessible tools for chemists and biologists to accelerate design of functional polymers and future nanomedicines.