AI Predicts Enzyme Function More Accurately

A team of researchers from the University of Illinois Urbana-Champaign has developed a new artificial intelligence (AI) tool called CLEAN that can predict the functions of enzymes based on their amino acid sequences. According to the researchers, this tool can predict the activity of enzymes even when they are unstudied or poorly understood, outperforming existing state-of-the-art tools in terms of accuracy, reliability, and sensitivity. The tool could significantly advance research in various fields, including genomics, chemistry, medicine, and pharmaceuticals.

Huimin Zhao, a chemical and biomolecular engineering professor at the university and the leader of the study, explained that the tool works by leveraging the language of proteins to predict their activity, just as ChatGPT uses data from written language to create predictive text. Zhao further said that when working with a new protein sequence, almost every researcher wants to know right away what the protein does. In addition, the tool will help researchers quickly identify the proper enzymes needed for the synthesis of chemicals and materials in various applications, such as biology, medicine, and industry.

Previous computational tools that try to predict enzyme functions attempt to assign an enzyme commission number that indicates what kind of reaction an enzyme catalyzes. However, these tools don’t work as well with enzymes that are less studied or uncharacterized or with enzymes that perform multiple jobs, according to Zhao. On the other hand, CLEAN uses a new deep-learning algorithm called contrastive learning to predict enzyme function, which works much better than other AI tools used by others, Zhao added.

The researchers verified their tool experimentally with both computational and in vitro experiments. They found that the tool not only predicted the function of previously uncharacterized enzymes, but also corrected mislabeled enzymes and identified enzymes with two or more functions.

The researchers plan to make CLEAN accessible online for other researchers who want to characterize an enzyme or determine whether an enzyme could catalyze a desired reaction. They also plan to expand the AI behind CLEAN to characterize other proteins, such as binding proteins, and further develop the machine-learning algorithms to point to a proper enzyme for a desired reaction.