Researchers at the University of Michigan have created the largest open-access database of chemical reactions to date, a resource that could significantly accelerate the development of new medicines and improve the performance of artificial intelligence systems used in chemistry.

The study was published in the Journal of the American Chemical Society.

At the heart of the project are more than 50,000 carefully designed chemical experiments. These experiments cover thousands of combinations of reagents and reaction conditions aimed at studying the formation of carbon–nitrogen bonds, which are key structural components of most modern pharmaceutical compounds.

The researchers note that drug development has traditionally relied on extensive trial and error, requiring thousands of experiments, considerable time, and the use of rare and expensive metals as catalysts. Although artificial intelligence has become increasingly important in chemistry, its effectiveness remains limited by the availability and quality of experimental data, which has long been insufficient.

The new database is intended to address this challenge. It has been made publicly available through the Open Reaction Database platform and is already considered the largest structured collection of chemical reaction data currently accessible to researchers.

According to the study’s lead investigator, Tim Cernak, the project took more than a decade to develop. While it currently covers only a small fraction of all possible chemical conditions, it is already capable of revealing patterns that would be difficult to identify through conventional research approaches.

One of the study’s key findings was a comparison of the performance of different catalysts, including palladium, nickel, and copper. Palladium is widely used in pharmaceutical synthesis, but its supply depends on a limited number of producing countries, creating potential vulnerabilities in global supply chains.

The researchers found that, in certain reactions, nickel can perform as effectively as palladium, while copper sometimes demonstrates comparable efficiency. These findings could pave the way for greater use of more abundant and geographically diverse materials in pharmaceutical manufacturing.

The team also identified unexpected chemical patterns, including the formation of highly reactive intermediate molecules at lower temperatures than previously believed possible. According to the researchers, such discoveries would have been difficult to achieve without analyzing large volumes of standardized experimental data.

The authors emphasize that combining extensive experimental datasets with artificial intelligence has the potential to transform drug discovery by making it faster, more cost-effective, and less dependent on scarce materials.