Chemists develop AI tool to convert human ingenuity into life-enhancing enzymes

Chemists develop AI tool to convert human ingenuity into life-enhancing enzymes Lisa Lock Scientific Editor Andrew Zinin Chief Editor Many have voiced concerns that artificial intelligence will take jobs away from humans. But assistant professor of chemistry Zhongyue (John) Yang, along with his former doctoral student Qianzhen Shao (now a postdoc at Stanford) and other team members, has created an AI tool that aims to elevate the work that scientists do—not replace them. MutexaGPT can speed up the "customization" of enzymes that drive countless chemical reactions essential to our daily lives. Engineered enzymes synthesize new medications, preserve nutrient-rich foods and help recycle polymers to reduce plastic waste. MutexaGPT takes plain-English, intuition-driven requests as input and leverages large language model agents to elicit missing information, build physics-based models and convert the results into actionable design proposals. The study is published in Nature Computational Science. "I know there are increasing concerns about AI replacing humans in the workforce, and I share them," Yang said. "However, with MutexaGPT, we are not using AI to get things done more efficiently without people. We are showing that with AI you can translate human intuition, which is often imaginative and incredible, to create a workable plan." He noted that enzymes, which are specialized proteins that speed up chemical reactions, can be viewed as nature's catalysts. Each enzyme consists of hundreds of amino acids, and changing even a single amino acid, which is called a mutation, can lead to substantial change in its function. "Now comes the major challenge, as no one knows how to predict mutations that alter genetic code without relying heavily on trial and error in industry," Yang said. "For the enzyme engineering market, which yields hundreds of billions of dollars in revenue each year, this simply means large amounts of super expensive testing and resource costs. MutexaGPT aims to change this paradigm by enabling researchers to use natural language to explore, predict and design enzyme mutations, drawing on artificial intelligence and fundamental principles of chemistry and physics." Among the tasks MutexaGPT can assist with is cold-active enzyme technology—engineering enzymes into versions that maintain high catalytic competence at lower temperatures. This is vital for cold-temperature processes like lactose-free milk and food preservation that ensures retention of heat-sensitive nutrients. Another practical application is the manufacture of laundry detergents that break down complex stains without the need for hot water. Similarly, designing enzymes with greater catalytic efficiency and selectivity makes pharmaceutical manufacturing more sustainable by reducing energy consumption, chemical waste and reliance on harsh reaction conditions. As for the name of this tool, Mutexa is short for "Alexa for Mutant." Yang said, "The name reflects our original vision of creating an Amazon Alexa–like intelligent assistant for designing enzyme mutants. "Just as Alexa helps people accomplish everyday tasks through natural-language conversations, MutexaGPT aims to help scientists navigate the complex world of enzyme mutations through an intuitive, conversational interface. The 'GPT' highlights the integration of large language models, which allow researchers to communicate with the system in natural language while connecting their questions to sophisticated computational tools for enzyme engineering." Yang, who grew up in China, was trained as a pure computational chemist while earning his Bachelor of Science from the Chang Po-ling Program of Chemistry at Nankai University. He then enrolled at UCLA in 2013, where he completed his doctorate in four and a half years. He had a postdoctoral fellowship at MIT prior to joining the Arts and Science faculty in 2020. Since then, he has been working to develop Mutexa and conduct experiments on its efficacy. "For researchers who work on enzyme engineering and design, they often develop their own intuition into how things work," Yang said. "They can have the principle but might not know how to quantify it," Yang said. "We can translate their principles into actual designs with MutexaGPT. I am excited to envision a future in which scientists can engineer enzymes for specific applications as readily as they consult an AI assistant, accelerating innovations in medicine, sustainable manufacturing and everyday consumer products." Publication details Qianzhen Shao et al., MutexaGPT: an intuition-to-design translator for physics-based enzyme engineering, Nature Computational Science (2026). DOI: 10.1038/s43588-026-01049-y Journal information: Nature Computational Science Provided by Vanderbilt University