CHAMPAIGN, Ill. — By combining artificial intelligence with automated robotics and synthetic biology, researchers at the University of Illinois Urbana-Champaign have dramatically improved performance of two important industrial enzymes — and created a user-friendly, fast process to improve many more.
Led by Huimin Zhao, a professor of chemical and biomolecular engineering at the U. of I., the team reported its findings in the journal Nature Communications.
“Enzymes have been increasingly used in energy production, in therapeutics, even in consumer products like laundry detergent. But they are not as widely used as they could be, because they still have limitations. Our technology can help address those limitations efficiently,” said Zhao, who also is affiliated with the Carl R. Woese Institute for Genomic Biology at the U. of I.
Enzymes are proteins that carry out specific catalytic functions that drive many biological processes. Those seeking to harness enzymes to advance medicine, technology, energy or sustainability often run into roadblocks involving an enzyme’s efficiency or its ability to single out a desired target amidst a complex chemical environment, Zhao said.
“Improving protein function, particularly enzyme function, is challenging because we don’t know exactly what kinds of mutations we should introduce — and it’s usually not just a single mutation; it’s a lot of synergistic mutations,” Zhao said. “With our model of integrating AI and automated synthetic biology, we offer an efficient way to solve that problem.”
Zhao’s group previously reported an AI model to predict an enzyme’s function based on its sequence. In the new paper, the researchers take their AI a step farther: predicting what changes to a known protein would improve its function.
“In a typically sized enzyme, the possible number of variations is larger than the number of atoms in the universe,” said Nilmani Singh, the co-first author of the paper. “So we use the AI method to help us create a relatively small library of potentially useful variant combinations, instead of randomly searching the whole protein sequence.”
However, training and improving an AI model is more than just code; it requires a lot of input, data and feedback. So the Illinois team coupled their AI models with the automated capabilities offered by the iBioFoundry, a center at the U. of I. dedicated to quick, user-friendly engineering and testing of biological systems ranging from enzymes to whole cells. Zhao directs the iBioFoundry, which is supported by the National Science Foundation.
In the new paper, the researchers lay out their process: First, they ask the AI tool how to improve a desired enzyme’s performance. The AI tool searches datasets of known enzyme structures and suggests sequence changes. The automated protein-building machines at the iBioFoundry produce the suggested enzymes, which are then rapidly tested to characterize their functions. The data from those tests are fed into another AI model, which uses the information to improve the next round of suggested protein designs.
“It’s a step toward a self-driving lab: a lab that designs its own proteins, makes the proteins, tests them and makes the next one,” said Stephan Lane, the manager of the iBioFoundry and co-first author. “The designing and learning is done by an AI algorithm, and the building and testing is done by robotics.”
Using this method, the team produced variants of two key industrial enzymes with substantially improved performance. One enzyme, added to animal feed to improve its nutritional content, increased its activity by 26 times. The other, a catalyst used in industrial chemical synthesis, had 16 times greater activity and 90 times greater substrate preference, meaning it was far less likely to grab molecules that were not its target.
“We described two enzymes in the paper, but it’s truly a generalized approach. We only need a protein sequence and an assay,” Zhao said. “We want to try to apply it to as many enzymes as possible.”
Next, the researchers plan to continue improving their AI models and upgrade equipment for even faster, higher-throughput synthesis and testing. They also have developed a user interface, enabling the system to run with a simple typed query. Their aim is to offer their method as a service for other researchers seeking to improve enzymes and speed drug development and innovations in energy and technology.
“For the user interface, the motivation is to allow people with different backgrounds to use the tool,” said graduate student Tianhao Yu, a coauthor of the paper. “If an experimental scientist doesn’t know how to run Python programs, then they can use our interface to help them run the program. They just need to use English to describe their needs, and it will automatically run.”
The National Science Foundation and the U.S. Department of Energy supported this work.