Creating custom enzymes is a grand challenge in protein science. Our latest research, appearing today in Science and recently across multiple preprints [1,2], showcases how AI-driven protein design can now be used to generate efficient protein catalysts with complex active sites tailored for specific chemical reactions.
What is enzyme design?
Enzymes are proteins that accelerate chemical reactions under mild conditions, far outperforming what even the best human chemists can achieve. Found in all living cells, some natural enzymes have been repurposed for biofuel and pharmaceutical production. They are even present in most laundry detergents, where they chew apart oil stains. Billions of dollars worth of industrial enzymes are produced and sold globally each year.
Generating custom enzymes that speed up important chemical reactions – whether building up pharmaceutical compounds or breaking down microplastics – would allow us to harness nature’s efficiency without relying on harsh solvents or fossil fuels. However, traditional enzyme engineering has relied on modifying existing proteins rather than building them from scratch.
“Traditional enzyme design is like buying a suit from a thrift store: the fit will probably be a little off. With AI, we can now tailor-make enzymes to ensure a perfect fit for every step of the reaction,” explained Sam Pellock, a co-lead author of the Science study and acting instructor in the lab.
Building new serine hydrolases
Sam, Anna, and Kiera led a project focused on accelerating a specific chemical reaction: breaking ester bonds. The enzymes they designed, called serine hydrolases, are unlike any found in nature. Through in silico modeling and lab validation, the team confirmed that these enzymes effectively bind and cleave ester compounds as intended.
“We tested our AI-designed enzymes in the lab and found they can be quite efficient,” said Anna Lauko, a co-lead author and recent PhD graduate. “There is still room for improvement because these chemical transformations are complex, but I’m thrilled by what we can now accomplish with the latest generation of design tools.”
Over 300 computer-generated proteins were tested in the lab. Of these, a subset showed reactivity with chemical probes, indicating successful installation of an activated catalytic serine. Through iterative rounds of design and screening, the team ultimately identified several highly efficient catalysts, including ones with activity levels that far exceed prior computationally designed esterases. Structural analysis revealed that the designed enzymes closely matched their intended architectures, with crystal structures deviating by less than 1 Å from their computational models. These results highlight the efficiency of integrating deep learning-based protein design with a novel assessment tool to evaluate catalytic preorganization across multiple reaction states.
A greener future
This work builds on other recent AI breakthroughs in our lab, including the development of a deep learning tool for modeling protein-small molecule conformational ensembles. This tool was used to create retroaldolase enzymes with considerably higher catalytic efficiencies than any pre-deep learning design for this reaction, as well novel enzymes that harness metal ions for catalysis with orders of magnitude higher catalytic efficiencies than previously designed metallohydrolases. This work is rapidly expanding the possibilities of enzyme design.
Kiera Sumida, a co-lead author of the Science study and a PhD trainee, is already applying these methods to tackle plastic degradation, demonstrating the broad potential of this approach. “I believe we’re not far from having custom enzymes that will help create a greener economy,” she said.
This work was made possible through support from the US National Institutes of Health, Defense Threat Reduction Agency, Howard Hughes Medical Institute, Washington Research Foundation, Open Philanthropy, The Audacious Project, Microsoft, and Schmidt Futures, with experimental support from the US Department of Energy.
Original Publication
Authors: Anna Lauko, Samuel J. Pellock, Kiera H. Sumida, Ivan Anishchenko, David Juergens, WoodyAhern, Jihun Jeung, Alex Shida, Andrew Hunt, Indrek Kalvet, Christoffer Norn, Ian R.Humphreys, Cooper Jamieson, Rohith Krishna, Yakov Kipnis, Alex Kang, Evans Brackenbrough, Asim K. Bera, Banumathi Sankaran, K. N. Houk, David Baker; Computational design of serine hydrolases; Science 2025
