South SFV Today

Enzymes play a vital role in various biological and industrial processes, from aiding digestion to enhancing the effectiveness of drugs. Scientists have long sought to develop new enzymes for diverse applications, including environmental protection and drug production. Now, a team of bioengineers and synthetic biologists has introduced a computational workflow that could revolutionize enzyme design.

The team's work is detailed in a recent paper published in Nature Communications. Michael Jewett, a professor of bioengineering at Stanford University and senior author of the study, explained the innovation: “We’ve developed a computational process that allows us to engineer enzymes much faster because we don’t have to use living cells to produce the enzymes.” Instead, machine learning models predict highly active designer enzymes based on mutated DNA sequences modeled on computers.

Traditionally, scientists relied on existing natural enzymes and made iterative changes using genetically modified cells in laboratories. This method often required thousands of iterations to achieve desired results. Jewett highlighted the advantage of their approach: “Rather than having to run 10,000 chemical reactions to iteratively improve enzyme activity, we can use machine learning models to predict highly active variants.”

The field known as "directed evolution" has been around for decades but applying machine learning accelerates this process significantly. The team’s workflow enables enzyme synthesis and testing without using living organisms. As proof of concept, they synthesized a small-molecule pharmaceutical with improved yield and demonstrated potential applications across multiple therapeutics.

Jewett is now seeking partnerships with pharmaceutical companies to further develop this model. He envisions expanding machine learning models for various chemical reactions beyond amide bond formation studied in this paper.

Despite these advancements, challenges remain due primarily to data scarcity. Jewett acknowledged the issue: “High-quality, high-quantity functional data remains a challenge.” Generating sufficient data for chemical reactions is slow but necessary as reliance on AI grows.

This research involved contributions from Grant M. Landwehr, Jonathan W. Bogart, Carol Magalhaes, Eric G. Hammarlund, and Ashty S. Karim at Northwestern University. Funding was provided by several organizations including NCI Cancer Center and National Institutes of Health.

###