(191e) Closed-Loop Discovery of Photostable Light-Harvesting Molecules

The design of light-harvesting molecules is limited by the challenge of simultaneously requiring high levels of photoabsorption with minimal amounts of photodegradation. In this talk, I present a closed-loop workflow that combines automated synthesis, photophysical characterization, and AI-guided prediction methods to discover new organic light-harvesting molecules with optimized photostability. This strategy is applied to a diverse chemical space of 2,200 oligomers comprised of electron donor, pi-bridge, and electron acceptor design motifs, resulting in the discovery of exceptional light-harvesting candidates while only requiring synthesis of <1.5% of the total chemical space. A Bayesian optimization framework is used to efficiently guide the search through a large molecular space using key physicochemical descriptors while maintaining a customizable tradeoff between exploitative and explorative sampling. Candidate molecules suggested by the AI framework are then prepared via automated synthesis using a “Lego-like” molecular building block approach based on Suzuki cross-coupling. Following synthesis, the physical properties of candidate molecules are characterized using an automated workflow, and experimental results are passed back to the Bayesian optimization scheme for subsequent rounds of closed-loop discovery. Regression models are trained to connect molecular descriptors with experimental photostability, uncovering a first principles understanding of the molecular determinants of photostability in which the triplet excited state manifold plays a critical role. Collectively, these results demonstrate that closed-loop discovery approaches lead to new fundamental chemical insights, while serving as a practical means for discovering new and efficient light-harvesting molecules to make organic photovoltaics a commercial reality for next-generation energy applications.