Scientists in Wisconsin have invented fully automated robots driven by artificial intelligence that search for millions of protein sequences, making the long process of protein engineering much faster. 

The team's invention improves efficiency in a relatively new field with huge potential: The protein engineering market is predicted to reach a value of $3.9 billion by 2024. With the right protein sequence, researchers could find cures for cancer, vaccines for any virus or even a solution to the world energy crisis.

"The central focus of our lab is using ideas from machine learning and artificial intelligence to try to accelerate this process of protein engineering," co-inventor Phil Romero told The Academic Times. The patent application was published by the U.S. Patent and Trademark Office on March 11.

Romero, who is also an assistant professor at the University of Wisconsin-Madison, started working in the field of protein engineering as a graduate student. At the time, he was working with Nobel laureate and chemist Frances Arnold. After studying traditional, slower methods to create new proteins, Romero "became very intrigued by this idea of using machine learning" to map the sequences of proteins. AI and robots accelerate the process, allowing researchers to focus their time on the outcomes and tinker with difficult chemical reactions instead of "searching randomly in this massive space," he said. 

The number of potential sequences for new proteins is mind-boggling. Proteins are formed by a chain of amino acids in a particular order, and this sequence encodes a protein with a unique function. Since there are 20 possible natural amino acids, a protein on the small end of the spectrum has 20 to the power of 100 possible sequences — a number "much larger than the number of atoms in the universe or seconds of age in the universe," Romero said. "It's absolutely enormous." The main challenge of protein engineering is "figuring out how to navigate this space of possibilities to find the few very rare sequences that might do something interesting," he said. 

The team's invention uses microfluidic technology as a tool to help search potential sequences. Microfluidic chips hold tiny amounts of fluids that allow scientists to watch biological reactions of small molecules or chemicals. This technique lets the researchers search through tens of millions of enzymes and variations instead of a few thousand at a time.

The AI-driven robots invented by Romero and his team are able to map sequences of amino acids and design new experiments on their own. The autonomous engineers, often called "self-driving labs," have two distinct parts. The first is an intelligent agent that analyzes sequence data to create new tests. It then sends these tests over to a robot, which can carry out the experiments by assembling genes, expressing proteins and running biochemical assays. The robots then "spit back the data points to the intelligent agent, who refines its understanding of the system to design" even more experiments, and the cycle repeats, Romero explained. 

This fully autonomous system can run with zero human intervention. In fact, it's so independent that Romero said it was "kind of fun seeing this robot go through the pandemic. There [were] at least two instances where the lab actually shut down because of a COVID infection on the floor, and the robot just keeps chugging away" through nights and weekends. The only intervention the system needs are reagents, added every few days, for a continuous process of protein engineering — and a robot technician, every once in a while.

The team paired with a company called Strateos that runs a "robotic cloud lab" in Palo Alto. This new-age laboratory is full of robots that perform experiments for researchers around the country. Romero's lab sends reagents to the robots in California, which execute the experiments and then return the data back to the protein engineering lab in Wisconsin. 

This partnership has worked out "really well because Strateos has a team of robotics engineers full-time to troubleshoot things as they break," Romero said. Having an on-site robots engineer is helpful when pieces in the robots get miscalibrated over time or programs crash, he added. Strateos' specialized knowledge is a benefit, he said, because, "It can be kind of tedious" for the researchers to try and solve mechanical issues themselves. And as the program grows, it will be much more efficient for one robot specialist to fix 50 robots if they are in the same room. 

Romero said he is "very excited about the potential of machine learning in protein engineering." He "imagines all sorts of crazy things for the future," such as "filling labs with these autonomous biological engineers and letting them do their thing, and coordinating their experiments in real-time through the cloud." 

The application for the patent, "Systems and methods for fully automated protein engineering," was filed Sept. 11, 2020, to the U.S. Patent and Trademark Office. It was published March 11, 2021 with the application number 17/018274. The earliest priority date was Sept. 11, 2019. The inventors of the pending patent are Philip Romero, Bennett Bremer, and Jacob Rapp, The University of Wisconsin-Madison. The assignee is Wisconsin Alumni Research Foundation.

Parola Analytics provided technical research for this story.