Scientists say they have developed a “self-driving” lab that uses artificial intelligence (AI) and automated systems to provide in-depth analyses of catalytic reactions used in chemical research and pharmaceutical manufacturing. The new tool, called Fast-Cat, can provide more information in five days than is possible in six months of conventional testing, claim the researchers, who add that at issue are the yield and selectivity of chemical reactions in the presence ligands.

The team published its study “Autonomous Reaction Pareto-Front Mapping with a Self-driving Catalysis Lab” in Nature Chemical Engineering.

“Ligands play a crucial role in enabling challenging chemical transformations with transition metal-mediated homogeneous catalysts. Despite their undisputed role in homogeneous catalysis, discovery and development of ligands have proven to be a challenging and resource-intensive undertaking,” write the investigators.

Self-driving catalysis

“Here, in response, we present a self-driving catalysis laboratory, Fast-Cat, for autonomous and resource-efficient parameter space navigation and Pareto-front mapping of high-temperature, high-pressure, gas–liquid reactions. Fast-Cat enables autonomous ligand benchmarking and multi-objective catalyst performance evaluation with minimal human intervention. Specifically, we utilize Fast-Cat to perform rapid Pareto-front identification of the hydroformylation reaction between syngas (CO and H2) and olefin (1-octene) in the presence of rhodium and various classes of phosphorus-based ligands.

“By reactor benchmarking, we demonstrate Fast-Cat’s knowledge scalability, essential to fine/specialty chemical industries. We report the details of the modular flow chemistry platform of Fast-Cat and its autonomous experiment-selection strategy for the rapid generation of optimized experimental conditions and in-house data required for supplying machine-learning approaches to reaction and ligand investigations.”

Fast Cat Hardware
Image shows the fully automated hardware of Fast-Cat, where the precursors are continuously delivered to the Fast-Cat flow reactor, catalytic reactions are rapidly conducted, and the reaction products are analyzed online to guide Fast-Cat’s decision-making process. [Milad Abolhasani, PhD/North Carolina State University]
From an industry perspective, you want the highest possible yield and selectivity. Because the specific steps you take when conducting the catalytic reaction can influence both yield and sensitivity, industrial chemists spend a tremendous amount of time and effort trying to find the parameters necessary to achieve the most desirable reaction outcome.

“The problem is that conventional catalyst discovery and development techniques are time-, material- and labor-intensive,” says Milad Abolhasani, PhD, corresponding author of the paper and an associate professor of chemical and biomolecular engineering at North Carolina State University. “These techniques rely largely on manual sample handling with batch reactors, as well as human intuition and experience to drive the experimental planning. In addition to the material inefficiencies, this human-dependent approach to catalyst development creates a large time gap between performing the reaction, characterizing the product, and making a decision about the next experiment.

“That’s why we created Fast-Cat. We’re now able to better understand how a specific ligand performs in five days than was previously possible in six months.”

Completely autonomous

Fast-Cat is completely autonomous, using AI and automated systems to continuously run high-temperature, high-pressure, gas-liquid reactions,” explains Abolhasani. The autonomous technology also analyzes the output from each of these reactions to determine–with no human intervention–how different variables affect the outcome of each experiment.

Fast-Cat uses the results from all of the previous experiments it has run—both successes and failures—to inform which experiment it will run next.

“Fast-Cat’s AI is constantly evolving, learning from the experiments it has already conducted,” Abolhasani says.

Users let Fast-Cat know what ligands and precursor chemicals it has to start with, and then see how much it can learn over 60 experiments.

“We spent a lot of time fine-tuning Fast-Cat’s AI model to optimize its ability to provide the broadest possible understanding of how different parameters affect the selectivity and yield of catalytic reactions using a specific ligand,” continues Abolhasani. “We also spent a lot of time ensuring that Fast-Cat’s findings are scalable. Fast-Cat conducts its experiments with extremely small sample sizes. But if we want its findings to be relevant for practical use, we needed to know that Fast-Cat’s findings hold true for reactions conducted on the large scales that are relevant for industrial manufacturing.”

The researchers have made the software and hardware publicly available so that Fast-Cat can be used to facilitate additional research,” points out Abolhasani ([email protected]), who hopes that “other researchers can adopt this technology to accelerate catalysis discovery in academia and industry.”

Previous articleTaming the Immune System When Rare Diseases Strike
Next articleUsing Patient-Derived Organoids to Predict Treatment Response in Pancreatic Cancer