EAGER: ADAPT: AI Guided Design and Synthesis of Semiconducting Molecules

Understanding how atoms and molecules combine to form more complex molecules and materials is fundamental to chemistry. Yet, there are more ways that atoms can be arranged into molecules than there are actual atoms in the universe. Chemists typically rely on experience, literature accounts, and ad hoc criteria for designing and prioritizing molecules for specific applications, often resulting in a monumental effort.

Furthermore, when considering the synthesis of new molecules using tried or untried reactions, chemists must often speculate, relying on instinct rather than ground truths. The PIs will apply and extend several aspects of artificial intelligence (AI) to develop a general platform that will facilitate data-driven property prediction and synthesis of semiconducting materials, focusing on blue light-emitting materials.

As a result, this project will expedite the discovery of novel organic semiconductors that can be synthesized efficiently, with optimized properties for target applications. The inclusion of a diverse team of graduate students in this work from the PIs’ research groups will broaden participation and help create an AI-aware workforce in the context of chemistry and materials science.

The research focus will be on optimizing the design of blue light-emitting materials using AI. To carry out this objective, the project will proceed with two parallel experimental tracks connected by an AI platform. In the first track, data and computational models will be used to train AI machine learning and experimental design modules for molecular-pair inputs.

This will provide a workflow that is fully containerized, enabling the design of robust blue light-emitting molecules. In the second track, the focus will be on extending the potential chemical space that can be integrated into the first track design concept. This will dramatically increase the diversity of semiconducting materials that can be explored and provide a roadmap for how to make new, unexplored molecular frameworks with desired properties.

The project will incorporate concepts and techniques from high-dimensional sparse regression, machine learning with graph inputs, and discrete optimization. The resulting dual mode platform (property design/synthesis design) will provide an unprecedented level of prediction, making the design and manufacturing of materials a more efficient and automated process.

At the same time, novel statistical machine learning and experimental design algorithms are expected to emerge in addressing chemistry problems involving molecular pairs as inputs. This project also provides new opportunities for undergraduate and graduate student training in materials and computational chemistry.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.