Researchers have developed a self-driving chemistry lab that can autonomously search through hundreds of catalyst recipes and reaction conditions to identify faster, more selective, and more programmable ways to make important industrial chemicals. The work could accelerate catalyst discovery for industries ranging from pharmaceuticals and plastics to fuels and specialty chemicals.
The new platform, called Flex-Cat, combines robotics, high-pressure chemical reactors, automated product analysis and artificial intelligence to discover both high-performing catalysts and catalysts whose behavior can be "tuned" to make different products by changing reaction conditions.
"Catalysts are the hidden engines of the chemical industry," says Milad Abolhasani, ALCOA Professor and University Faculty Scholar in the Department of Chemical and Biomolecular Engineering at North Carolina State University and co-corresponding author of the study. "They help transform raw materials into the chemicals used to make plastics, medicines, fuels and many other products we rely on every day.
"But finding the right catalyst is extremely challenging. It is not enough to identify the right catalytic material. You also have to find the right temperature, pressure, and concentration for that catalyst to work efficiently and selectively. That creates a vast experimental search space."
Flex-Cat was designed to address that challenge. The autonomous system can prepare catalysts, run high-pressure reactions, analyze the products and use the results to decide which experiments to run next. In this study, the researchers used Flex-Cat to study hydroformylation, an industrially important reaction that converts simple chemical feedstocks into aldehydes – versatile building blocks used in the manufacture of plastics, surfactants, solvents, and other everyday materials.
A key challenge in hydroformylation is controlling which aldehyde isomer is produced. Isomers contain the same atoms but are arranged differently, giving them different chemical properties and industrial uses. In practical terms, chemists often want to steer the reaction toward one product or another, but doing so requires the right pairing of catalyst structure and reaction conditions. Researchers can direct Flex-Cat to automatically optimize the catalyst and conditions to maximize the yield of a single aldehyde isomer product.
"Conventional catalyst discovery is slow, expensive and heavily dependent on human intuition," Abolhasani says. "Flex-Cat turns that process into an autonomous learning cycle. The system makes a decision, runs the experiment, learns from the outcome and then chooses the next best experiment."
Over the course of the study, Flex-Cat performed 680 experiments using 16 chemically diverse phosphorus-based ligands, which are molecules that tune the behavior of the rhodium catalyst used in the reaction. The system carried out three autonomous optimization campaigns: one targeting the branched aldehyde product, one targeting the linear aldehyde product, and one designed to find catalysts that could switch between product types depending on reaction conditions.
The results were striking. Flex-Cat identified catalyst–condition combinations that improved catalyst activity by more than 2.5-fold, expanded the accessible range of product selectivity, and uncovered ligands that could be programmed to favor different products under different conditions.
"From a chemist's perspective, Flex-Cat is particularly exciting because it not only optimizes a desired reaction outcome," says Alexander Miller, professor of chemistry at the University of North Carolina at Chapel Hill and co-corresponding author of the study. "The platform has helped us understand how catalyst structure and reaction conditions work together to control selectivity. That gives chemists insights that can point the way toward designing even better catalysts in the future."
"This work is exciting because it connects autonomous discovery with the types of questions that matter in industrial catalysis," says Damon Billodeaux, Group Leader for Texas Process Innovation and Technology Manager for Olefins Stream at Eastman, which supported the research. "Being able to rapidly identify catalyst systems that are both high-performing and tunable could help accelerate the development of more efficient and flexible chemical manufacturing processes."
"That tunability is what makes this exciting," Abolhasani says. "Flex-Cat did not just find a good catalyst for one product. It found catalysts that behave almost like chemical dimmer switches; the same catalytic system can be pushed toward different products by changing the operating conditions."
The platform also revealed chemical design trends that can help guide future catalyst development. By analyzing the autonomously generated dataset, the researchers identified how ligand structure influenced product selectivity and flexibility, offering insight into why some catalysts were biased toward one product while others were more condition-programmable.
"This is not just about finding one catalyst," Abolhasani says. "The larger impact is that Flex-Cat gives us a general strategy for mapping complex catalytic systems, identifying promising regions quickly, and generating actionable knowledge for chemical process development."
The paper, " An Autonomous Lab for Data-Driven Homogeneous Catalysis ," is published in the open-access journal Nature Communications and was supported by the Eastman Chemical Company. Other NC State contributors are Jeff Bennet, postdoctoral researcher, and Ph.D. students Negin Orouji and Alireza Velayati. Eamon Reynolds and Matthew Porowski, Ph.D. students at UNC-Chapel Hill, and Manik Saha from Eastman, also contributed to the work.
