Autonomous Discovery of Functional Random Heteropolymer Blends through Evolutionary Formulation Optimization

While developing new polymers typically requires years of investigation, blending existing polymers offers a cost-effective strategy for creating new materials that meet specific requirements. Yet identifying functional polymer blends is often a laborious development process, complicated by the vast design space and non-additive nature of polymer properties, exacerbated by an often-limited understanding of structure-function relationships. To this end, we report an autonomous closed-loop platform with an evolutionary algorithm for the development of functional polymer blends. We focus on random heteropolymers (RHPs), which are gathering increasing interest as versatile materials with a range of promising applications. Using enzyme thermal stabilization as an objective, we identify blended compositions from combinatorial 96- or 192-dimensional spaces (with over 10^9 potential candidates) that exhibit emergent function and outperform all of their constituent polymers by an absolute margin of 26% retained enzyme activity. Our findings highlight the immense potential of leveraging autonomous closed-loop discovery platforms for polymer blend discovery, as well as the opportunity for materials discovery within the RHP blend space. The algorithmic goal of blend optimization also bears a strong resemblance to other formulation optimization problems that are pervasive in molecular and material discovery.