An enzyme library makes a wide range of high-purity amines from equal starting amounts
Researchers tested 175 enzymes and used sequence data and an AI system to identify catalysts for producing optically pure amines at gram scale.
Editorial illustration — not from the study.
Researchers combined iterative laboratory screening, sequence analysis and machine-learning-guided testing to examine 175 enzymes across chemically varied starting materials. They found that asymmetric reductive amination—the enzyme-catalysed joining of an aldehyde or ketone with an amine while controlling the product’s handedness—was widespread in this enzyme family even when the two starting materials were used in equal amounts.
The team assembled enzymes with broad activity, high specific activity and strong control over product handedness. Sequence analysis linked broad activity to characteristic enzyme features and identified residues associated with stereoselectivity. Models trained on the screening results also predicted enzyme activity for starting-material combinations that had not previously been tested. Selected reactions were run at preparative scale and produced gram quantities of optically pure amines.
What the enzymes could do
Across 175 imine reductases and reductive aminases, the researchers profiled reactions involving structurally diverse substrate panels. They found that asymmetric reductive amination at equimolar substrate concentrations was widespread across the enzyme family. Some selected catalysts combined broad substrate scope, high specific activity and excellent stereoselectivity.
Sequence-level analysis identified features associated with broad-scope enzymes and residues linked to stereoselectivity. Machine-learning models trained on the screening data successfully predicted enzyme activity for previously untested substrate combinations. Selected reactions were scaled to preparative batch size, yielding gram quantities of optically pure amine products.
Why the library matters
Chiral amines are useful building blocks in chemical synthesis, and enzymes that make them with controlled three-dimensional structure could provide a more systematic way to prepare varied products. The study suggests that useful activity is not limited to a small number of narrowly specialized enzymes: a data-guided library can help find catalysts that work across broader sets of starting materials while using those materials in equal amounts.
The sequence findings and activity predictions may also help researchers choose or design enzymes for new substrate combinations instead of testing every possibility experimentally. The work demonstrates this approach for the enzyme family and reaction panels studied, rather than establishing that it will work for all amines or chemical processes.
Evidence and open limits
The evidence comes from large-scale enzyme screening, sequence analysis, machine-learning predictions and selected preparative-scale reactions. The study directly reports activity across 175 enzymes and gram-scale production of optically pure products in selected reactions.
The abstract does not specify the exact number of products made at gram scale, their individual yields, the full substrate list, or the accuracy of the prediction models. It also does not establish how well the catalysts perform outside the tested substrate panels or whether the findings apply broadly to other enzyme families and manufacturing conditions.
Researchers report a nickel-iron catalyst that sustained water electrolysis for more than 3,400 hours at an industrial current density. The reported hydrogen production cost was $2.28 per kilogram, below the European Commission’s 2030 target.
Researchers developed Orbformer, an AI model that learns transferable descriptions of molecules from 22,000 structures, including molecules with breaking bonds. After fine-tuning, it reached chemical accuracy on established tests and more demanding bond-breaking and Diels–Alder reaction problems, while reducing the need to solve each new molecule from scratch.
A compact near-infrared device uses lead-free perovskite materials to screen alcohol and water content in commercial samples. Its circuit turns optical signals into five visual levels, covering alcohol concentrations of 30–60% and water contents of 45–100%.
