Artificial intelligence and the molecular machine join forces to generalize automated chemistry

Artificial intelligence, “building block” chemistry and a molecule-making machine have come together to find the best general reaction conditions for synthesizing chemicals important to biomedical and materials research – a discovery that could accelerate innovation and discovery of drugs, and to make complex chemistry automated and accessible.

With machine-generated optimized conditions, researchers at the University of Illinois Urbana-Champaign and collaborators in Poland and Canada doubled the average yield of a special, difficult-to-optimize type of reaction that bonds carbon atoms together in pharmaceutically important molecules. The researchers say their system provides a platform that can also be used to find general conditions for other classes of reactions and solutions to similar complex problems. They reported their findings in the journal Science.

“Commonality is critical to automation, and thus molecular innovation becomes accessible even to non-chemists,” said study co-leader Dr. Martin D. Burke, professor of chemistry at Illinois and the Illinois Carle College of Medicine, as well as a physician . “The challenge is that the haystack of possible reaction conditions is astronomical, and the needle is hidden somewhere in there. Using the power of artificial intelligence and the chemistry of building blocks to create feedback, we were able to shrink the haystack. And we found the needle.”

Automated machines for synthesizing proteins and nucleic acids such as DNA have revolutionized research and chemical production in these fields, but many chemicals important for pharmaceutical, clinical, manufacturing and materials applications are small molecules with complex structures, researchers say.

Burke’s group pioneered the development of simple chemical building blocks for small molecules. His lab has also developed an automated molecule-making machine that connects the building blocks to create a wide range of possible structures.

However, the general reaction conditions to make the automated process widely applicable remain elusive.

“Traditionally, chemists have customized the reaction conditions for each product they’re trying to make,” Burke said. “The problem is that this is a slow and very specialist-dependent process and very difficult to automate, as the machine will have to be optimized each time. What we really want are conditions that work almost every time, no matter what two things I’m trying to put together.”

An automated approach with generalized conditions could help standardize the way some products are made, addressing the issue of reproducibility, said Illinois postdoctoral researcher Vandana Rathore, co-author of the study.

Burke’s group teamed up with a group led by Bartosz A. Grzebowski of the Institute of Organic Chemistry of the Polish Academy of Sciences, as well as Alan Aspuru-Guzik’s group at the University of Toronto, both leaders in the use of artificial intelligence and machine learning to improving chemical synthesis. The team integrated AI with the molecular machine to provide real-time feedback to the machine learning system.

“To tell the good from the bad, you have to know something about the bad, but people only post the successes,” Grzybowski said. Published studies reflect conditions that are popular or convenient, not the best, so a systematic approach that includes diverse data and negative outcomes is needed, he said.

First, the team ran the entire matrix of possible combinations using building block chemistry through an algorithm to cluster similar reactions. The AI ​​then sent instructions entered into a machine at the Molecule Maker Lab, located at the Beckman Institute for Advanced Science and Technology in Illinois, to produce representative reactions from each cluster. Information from these reactions is fed back into the model; The AI ​​learned from the data and ordered more experiments from the molecular machine.

“We wanted to see two things: increase the yield and decrease the uncertainty for a wide range of reactions,” said Grzybowski, who is now at the Ulsan Institute of Science and Technology in South Korea. “This cycle continued without us having to intervene until the problem was resolved. It took 30 years to come up with the general terms for protein synthesis machines. It took us two months.”

The process identifies conditions that double the average yield of a challenging class of reactions called Suzuki-Miyaura heteroaryl coupling, which is critical to many biological and materials-related compounds.

“There are all kinds of combinations of building blocks that we didn’t even study in our AI training, but because the AI ​​had explored such a diverse space, it found good results even in these initially unexplored areas,” said Illinois graduate student Nicholas H. Angelo, co-author of the study.

The machine learning process described in the paper can also be applied to other broad areas of chemistry to find the best reaction conditions for other types of small molecules or even larger organic polymers, the researchers say.

“There are so many different classes of materials that we want to know, target and discover for different functional properties. The opportunity to extend this approach to other similar reaction chemistry, other types of carbon-carbon bonds, is exciting,” said study co-author Charles M. Schroeder, Illinois professor of materials science and engineering and chemical and biomolecular engineering and a Beckman Institute affiliate. .