A closed-loop transfer paradigm for physics-based functional molecule discovery. Credit: Nicholas Angello et al.
Artificial intelligence is a powerful tool for researchers, but it has a major limitation: it cannot explain how it reaches decisions, a problem known as the “AI black box.”
By combining AI with automated chemical synthesis and experimental validation, an interdisciplinary team of researchers at the University of Illinois at Urbana-Champaign opened the black box and discovered the chemical principles that the AI relied on to improve molecules for harvesting solar energy.
The result was a light-harvesting molecule that was four times more stable than the starting point, and also provided important new insights into what makes the molecule stable – a chemical question that has hindered materials development.
The interdisciplinary research team was co-led by Illinois chemistry professor Martin Burke, chemical and biomolecular engineering professor Ying Diao, chemistry professor Nicholas Jackson, and materials science and engineering professor Charles Schroeder, in collaboration with University of Toronto chemistry professor Alan Aspur-Guzik, who published their findings in the journal Nature.
“New AI tools are incredibly powerful, but when we open the hood and try to understand what they're doing, we're often left with nothing useful to work with,” Jackson said.
“For chemistry, this is incredibly frustrating. AI can help you optimize a molecule, but it can't tell you why it's optimal — what are the key properties, structure, and function. Through our process, we've identified what gives these molecules high photostability. We've turned the AI black box into a clear glass sphere.”
The researchers were motivated by the question of how to improve organic solar cells by making them based on thin, flexible materials, as opposed to the rigid, heavy, silicon-based panels that currently dot rooftops and fields.
“The biggest obstacle to commercializing organic photovoltaics is stability issues: high-performance materials degrade when exposed to light, which is undesirable for solar cells,” Diao said. “Organic photovoltaics can be manufactured and deployed in ways that silicon can't, and they can also convert heat and infrared radiation into energy, but stability has been an issue since the 1980s.”
The Illinois method, called “closed-loop transfer,” starts with an AI-guided optimization protocol called a closed-loop experiment: The researchers asked the AI to optimize the photostability of the light-harvesting molecules, Schroeder said.
The AI algorithm provided suggestions on what chemicals to synthesize and explore in multiple closed-loop synthesis and experimental characterization rounds. After each round, new data was fed back into the model to provide improved suggestions, with each round getting closer to the desired outcome.
Thanks to the building-block-like chemistry and automated synthesis developed by Burke's group, the researchers generated 30 new chemical candidates over five closed-loop experiments. The work was conducted in the Molecule Maker Lab at the University of Illinois' Beckman Institute for Advanced Science and Technology.
“The modular chemistry approach beautifully complements closed-loop experimentation: AI algorithms ask for new data that maximizes their learning potential, and our automated molecular synthesis platform can very quickly generate the new compounds needed. These compounds are then tested and the data is fed back into the model, making the model smarter again and again,” said Burke, who is also a professor at the Carle Illinois College of Medicine.
“Up until now, we have focused primarily on structure. Our automated modular synthesis is now moving into the realm of exploring function.”
Rather than simply ending the query with an AI-selected final product, as in a typical AI-driven campaign, the closed-loop forwarding process sought to further uncover hidden rules that would make the new molecule more stable.
While the closed-loop experiment was running, another set of algorithms continually observed the resulting molecules, developing models of their chemical signatures that predicted their stability in light, Jackson said. Once the experiment was finished, the models provided new hypotheses that could be tested in the lab.
“We use AI to generate hypotheses, test them and trigger new human-led discovery campaigns,” Jackson said.
“Now that we have some physical descriptors that make a molecule photostable, this makes the screening process for potential new chemical entities dramatically easier than blindly searching chemical space.”
To test their photostability hypothesis, the researchers investigated three structurally distinct light-harvesting molecules that share the chemical properties (specific high-energy regions) they identified, and found that choosing the right solvent increased the molecules' photostability by up to four-fold.
“This is a proof of principle of what is possible. We are confident that we can address other material systems as well, and the possibilities are limited only by our imagination. Ultimately, we envision an interface where researchers can input their desired chemical functions and an AI will generate hypotheses to test,” Schroeder said.
“This research was only possible thanks to the multidisciplinary team and the talent, resources and facilities at the University of Illinois, and our collaborators in Toronto. The five groups worked together to generate new scientific insights that would not have been possible if any of the subteams had been working in isolation.”
Further information: Nicholas Angello et al., “Closed-loop transfer enables artificial intelligence to generate chemical knowledge,” Nature (2024). DOI: 10.1038/s41586-024-07892-1. www.nature.com/articles/s41586-024-07892-1
Courtesy of University of Illinois at Urbana-Champaign
Citation: Breaking through AI's black box, team discovers key chemistry for solar energy and beyond (August 28, 2024) Retrieved August 29, 2024 from https://phys.org/news/2024-08-ai-black-team-key-chemistry.html
This document is subject to copyright. It may not be reproduced without written permission, except for fair dealing for the purposes of personal study or research. The content is provided for informational purposes only.
