UNC Scientists Win $1 Million W.W. Keck Foundation Award to Decode the Secrets of AI-Designed Proteins
Gary Pielak, left, Kenan Distinguished Professor of Chemistry, Biochemistry and Biophysics, and Brian Kuhlman, Oliver Smithies investigator and Professor of Biochemistry and Biophysics, will focus their attention on AI-designed proteins, which can fold into beautiful, compact shapes and sometimes act as enzymes—the tiny “chemical robots” that speed up the reactions that make life possible. But there’s a catch: AI-designed proteins often behave differently from natural ones, even when they look similar on paper.
November 23, 2025 I By Dave DeFusco
When you think of technology reshaping the future, you might picture robots, self-driving cars or smart devices. Some of the biggest breakthroughs, however, may come from a much smaller world—the world of proteins, which are the molecules that carry out almost every task inside cells. Two scientists at UNC-Chapel Hill, Gary Pielak and Brian Kuhlman, have just received a $1 million Science and Engineering Research Award from the W.M. Keck Foundation to study these molecules in a new way.
Their work centers on a surprising puzzle. Proteins designed by artificial intelligence—once considered almost science fiction—now exist in labs around the world. These AI-designed proteins can fold into beautiful, compact shapes and sometimes act as enzymes, the tiny “chemical robots” that speed up the reactions that make life possible. But there’s a catch: AI-designed proteins often behave differently from natural ones, even when they look similar on paper.
“Something unusual is going on inside these designed proteins,” said Pielak, Kenan Distinguished Professor of Chemistry, Biochemistry and Biophysics. “They’re often far more stable than proteins found in nature and, in many cases, their interiors look almost molten or unusually flexible. We want to understand why.”
Natural enzymes are marvels of evolution. They’re shaped over millions of years to be just stable enough to do their job, but not so rigid that they can’t move or adjust when they need to. This balance of structure and motion is what gives them their extraordinary speed and precision. AI-designed enzymes, however, don’t always follow nature’s rules. Many stay folded even at temperatures near boiling water. Others seem soft or flexible at their core. Yet, these “unnatural” behaviors come from a design process that learns from natural proteins.
Understanding these differences matters, not just for science but for industry. Enzymes could one day replace many chemical processes that rely on toxic metals or harsh solvents. If researchers can learn how to design enzymes to be as efficient as natural ones, it could transform the pharmaceutical and chemical industries with cleaner, greener technologies. Pielak and Kuhlman, Oliver Smithies investigator and Professor of Biochemistry and Biophysics, will explore the mystery from two angles.
First, they will study two AI-designed enzymes created in the laboratory of Nobel Prize-winning scientist David Baker. One of these proteins, called LuxSit-i, stays folded even at or above the boiling point of water. Early data suggests that it may have a soft, flexible center.
The second enzyme, Dad t2, works like a natural enzyme but is far less efficient. Pielak’s team will use a powerful technique called nuclear magnetic resonance spectroscopy (NMR) to look inside these proteins and measure how rigid or flexible they are, how stable they remain under heat and how well they perform their catalytic tasks.
“NMR lets us see proteins at the level of individual atoms,” said Pielak. “We can measure how each part of the molecule moves and how that movement relates to stability and, perhaps, function. It’s like watching the heartbeat of a protein.”
Second, the researchers will take a natural enzyme called adenylate kinase and redesign it using AI. They want to see whether the redesigned version becomes super-stable, like many AI-created enzymes, and whether that added stability comes with a molten or flexible core.
Comparing the redesigned enzyme with natural versions from cold-loving, moderate-temperature and heat-loving organisms will help reveal how evolution tuned these proteins for different environments.
“We’ll be testing whether AI pushes proteins into states that evolution never chooses,” said Kuhlman, “and, if so, why? What trade-offs is the machine making that nature avoids?”
If AI-designed enzymes are too soft or too rigid on the inside, that may explain why they often don’t match the incredible speed and precision of natural enzymes. Understanding this gap could help scientists design better molecules—ones that combine the stability of AI creations with the fine-tuned performance of natural proteins. The payoff could be huge. Enzymes are environmentally friendly: they work in water, avoid toxic metals and use less energy. Better designed enzymes could clean up industrial processes, unlock new therapies and create sustainable ways to build essential chemicals.
“We want to make the next generation of enzymes smarter, greener and more reliable,” said Pielak. “But to do that, we first have to understand how nature works and why AI sometimes breaks the rules.”
Pielak brings more than 40 years of experience probing protein stability and dynamics. He pioneered the use of in-cell NMR, revealing how proteins behave inside living cells, which is a far more chaotic environment than the test tube. Kuhlman is internationally recognized for his work in protein design, including the first design of an entirely new protein fold. Together, their expertise bridges evolution and computation, experiment and theory.
“The Keck Foundation recognized that we’re at a turning point,” said Kuhlman. “AI is advancing fast, but we’re missing key data about how these designed proteins behave. This project fills that gap. If we’re successful, we’ll identify ways that AI-based protein modeling can be improved to design proteins with important applications in medicine, industry and research.”