Closed-loop transfer enables artificial intelligence to yield chemical knowledge.

Area of Science:

• Materials Science
• Chemistry
• Artificial Intelligence

Background:

• Artificial intelligence (AI)-guided closed-loop experimentation optimizes functions but often acts as a black box, limiting chemical knowledge discovery.
• The potential of AI in uncovering novel chemical insights alongside optimization remains largely unexplored.

Purpose of the Study:

• To integrate closed-loop experiments with physics-based feature selection and supervised learning, termed closed-loop transfer (CLT).
• To achieve parallel optimization of objective functions and generation of chemical insights.
• To investigate factors influencing photostability in solution for light-harvesting donor-acceptor molecules.

Main Methods:

• Developed and applied closed-loop transfer (CLT), combining automated synthesis, experimental characterization, physics-based feature selection, and supervised learning.
• Explored a small fraction (1.5%) of the theoretical chemical space for efficiency.
• Validated the physics-informed photostability model using multiple experimental test sets and solvent tuning.

Main Results:

• CLT successfully yielded fundamental chemical insights into photostability, highlighting the importance of high-energy triplet states.
• Achieved significant optimization of objective functions with minimal exploration of chemical space.
• Demonstrated the generalizability of CLT through applications to additional materials systems.

Conclusions:

• Combining interpretable supervised learning with physics-based features in closed-loop discovery rapidly provides fundamental chemical insights.
• CLT offers a powerful strategy to augment closed-loop discovery processes for both optimization and knowledge generation.
• This approach enhances the discovery of new materials for organic electronics and other applications.