Bayesian optimization for multicomponent supramolecular systems

The diversity of noncovalent interactions makes the design space of multicomponent molecular systems highly complex. To efficiently explore supramolecular design space, data-driven strategies are needed. Here, we demonstrate a methodological framework for the targeted design of multicomponent molecular systems with noncovalent interactions using Bayesian optimization. Its effective applicability to supramolecular polymers is illustrated by three representative cases that reveal accelerated exploration of diverse multicomponent systems with a universal Bayesian optimization framework. The number of experiments required to arrive at optimal compositions is significantly reduced compared to random or uninformed sampling strategies, enabling the experimental study of high-dimensional design spaces. In this way, we can tune the formulation of intricate mixtures and achieve tailored macroscopic properties with minimal experimental effort. Our results show that Bayesian optimization is a general tool for developing multicomponent supramolecular systems with designed functionality.