If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)

ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.



Pulsed electrosynthesis plus machine learning equals bigger yields for adiponitrile

Data-driven reaction optimization may inspire more efficient paths to other chemicals

by Leigh Boerner
September 6, 2019 | A version of this story appeared in Volume 97, Issue 35


An electrochemical cell showing acrylonitrile combining with water and electrons to make adiponitrile and OH<sup>–</sup> ions at the cathode, and the oxidation of OH<sup>–</sup> ions to water and oxygen at the anode. On the left side of the cell, electrolytes flow through the input and output.
Credit: Modestino lab
In the Modestino group’s electrochemical cell, acrylonitrile (AN) molecules combine to make adiponitrile (ADN) at the cathode, while water and oxygen form at the anode. Electrolytes flow through the input and output ports (left).

Adiponitrile is kind of a big deal. One of the most produced chemicals worldwide by volume, it’s an important precursor for making the compound nylon 6,6, commonly used in carpeting and textiles. But the methods for manufacturing adiponitrile could do with some improvement.

Currently, industry synthesizes the chemical via two pathways. The less-often-used, yet theoretically more environmentally friendly, route is one that is driven by electrochemistry. The problem with this route is that it is inefficient and can create unwanted by-products. Now Miguel Modestino and coworkers at New York University have increased both the yield and the selectivity of this reaction by pulsing the current in an electrosynthesis system and optimizing their reaction conditions with machine learning (Proc. Natl. Acad. Sci. U.S.A.2019, DOI: 10.1073/pnas.1909985116).

Normally, when scientists electrochemically synthesize adiponitrile, they use a constant electrical current to drive the reaction. But that constant current reduces the reactant acrylonitrile at the system’s electrode, where it can react with water to form an unwanted by-product, propionitrile. Pulsing the current in millisecond bursts, though, allows time for the reduced form of acrylonitrile to diffuse away from the electrode, says graduate student Daniela Blanco, an author of the new study. And this diffusion leads to the boost in adiponitrile yield and reaction selectivity.

Scheme showing the proposed cathodic pathway for the electrochemical reaction of acrylonitrile to adiponitrile.
During the electrosynthesis of adiponitrile (ADN), acrylonitrile (AN) is reduced at the system's cathode and then combines with water, electrons, and an unreduced version of itself to yield the final product.

The researchers quickly realized, however, that small changes to the pulse rate translated to large, unpredictable changes in reaction activity, making the system’s conditions difficult to optimize. This is when they turned to machine learning, Modestino explains. Using the computer software Matlab, the team created a type of machine-learning algorithm called a neural network and trained it using data from 16 experiments to predict better reaction conditions for making adiponitrile. When they applied the reaction conditions outputted by the neural network to their electrosynthesis setup, the researchers were able to increase the selectivity of the reaction 325% and boost production of adiponitrile by 30% over standard electrosynthesis—the largest reported improvement since the discovery of this reaction more than 50 years ago.

“Data-driven optimization is not common in electrochemical synthesis, and I think this paper will likely inspire many others to take similar approaches,” says Karthish Manthiram, an electrochemical engineer at the Massachusetts Institute of Technology.

Artificial intelligence experts unaffiliated with the new study expressed concern to C&EN over the small number of data points used to train the team’s neural network. Machine-learning algorithms typically need large, costly, high-quality training data sets to make accurate predictions. Modestino sees the small number of data points his team used as a positive, however, because the researchers were still able to optimize the reaction with a limited amount of data. “In our field, it’s so expensive to run new experiments,” Modestino says. Getting more from the data you already have could make a big impact, he says.

Modestino, Blanco, a former graduate student formed the start-up Sunthetics to focus on reducing greenhouse gas emissions from traditional chemical processes by switching to electrochemically powered ones. While they’re focusing on the adiponitrile reaction for now, they hope to use the same techniques to improve other chemical manufacturing processes.


This article has been sent to the following recipient:

Chemistry matters. Join us to get the news you need.