AI robots carry out chemical synthesis research

In a study published in Nature, the research team demonstrated how mobile robots employing AI-driven decision-making could carry out exploratory chemistry tasks at a human level but with significantly greater speed.

Standing 1.75 metres tall, the robots were designed by the Liverpool team to address three key challenges in exploratory chemistry: performing chemical reactions, analysing the results, and determining the next steps based on collected data.

The robots worked together to tackle problems in three areas of chemical synthesis: structural diversification chemistry (important for drug discovery), supramolecular host-guest chemistry, and photochemical synthesis.

The study found that the AI-enabled robots made decisions comparable to those of a human researcher but on a far shorter timescale, completing in seconds tasks that could take hours for a person.

Professor Andrew Cooper, from the University’s Department of Chemistry and Materials Innovation Factory, who led the project, explained: “Chemical synthesis research is time-consuming and expensive, both in the physical experiments and the decisions about what experiments to do next, so using intelligent robots provides a way to accelerate this process.

“When people think about robots and chemistry automation, they tend to think about mixing solutions, heating reactions, and so forth. That’s part of it, but the decision-making can be at least as time-consuming. This is particularly true for exploratory chemistry, where you’re not sure of the outcome. It involves subtle, contextual decisions about whether something is interesting or not, based on multiple datasets. It’s a time-consuming task for research chemists but a tough problem for AI.”

Decision-making is a core challenge in exploratory chemistry. For instance, a researcher might conduct multiple trial reactions and then scale up only those with promising yields or interesting products. For AI, determining what is ‘interesting’ depends on multiple factors, such as the novelty of a product or the cost and complexity of the synthesis.

Dr Sriram Vijayakrishnan, a former University of Liverpool PhD student and Postdoctoral Researcher in the Department of Chemistry, who led the synthesis work, commented: “When I did my PhD, I performed many of the chemical reactions manually. Often, collecting and analysing the data took as long as setting up the experiments. This data analysis challenge becomes even more significant when you automate chemistry, as you can end up overwhelmed by data.

“We addressed this by developing an AI logic for the robots. This processes analytical datasets and makes autonomous decisions – for example, whether to proceed to the next reaction step. This decision is essentially instantaneous. If the robot analyses data at 3:00am, it will decide by 3:01am which reactions to progress. In contrast, a human chemist might take hours to review the same data.”

Professor Cooper added: “The robots have less contextual breadth than a trained researcher, so in its current form, it won’t have a ‘Eureka!’ moment. However, for the tasks we gave it, the AI logic made decisions that were comparable to those of a synthetic chemist across three different chemistry problems, and it did so almost instantaneously. There is also significant potential to enhance the AI’s contextual understanding, such as by integrating large language models to directly connect it with scientific literature.”

Looking ahead, the Liverpool team aims to use this technology to discover chemical reactions relevant to pharmaceutical drug synthesis and to develop new materials for applications such as carbon dioxide capture.

Although this study utilised two robots, the system is scalable, with no inherent limit to the size of robotic teams. This approach could therefore be expanded to suit the largest industrial laboratories.

This research builds on the University’s earlier work on the world’s first “mobile robotic chemist,” reported by Professor Cooper’s team in 2020 (Nature, 2020, 583, 237). That robot conducted nearly 700 catalysis experiments over eight days, working continuously.

This paper follows the recent award of the 2024 Nobel Prize in Chemistry to researchers who used AI to predict protein structures, reflecting the rapid adoption of AI in scientific research.

The project received funding from the Leverhulme Trust, the European Research Council, and the Engineering and Physical Sciences Research Council.