Vanderbilt Team Launches MutexaGPT for Enzyme Design
The AI tool turns plain-English requests into physics-based enzyme engineering proposals, aiming to replace trial-and-error methods.
Updated on Nov. 3, 2026 in Biotech
Vanderbilt Assistant Professor Zhongyue (John) Yang and his team created MutexaGPT, an AI tool that takes plain-English requests and uses large language model agents to build physics-based models for enzyme design. The tool converts its results into actionable design proposals.
Why it matters
Enzyme engineering is a multibillion-dollar field in which mutation prediction has relied heavily on trial and error. MutexaGPT is intended to change that paradigm, which could matter for Nashville's research institutions and the biotech sector.
The enzyme engineering market generates hundreds of billions of dollars in revenue each year, and each enzyme is built from hundreds of amino acids whose single-residue changes are called mutations. The source does not report accuracy or benchmark results for MutexaGPT against existing methods.
The players
Zhongyue (John) Yang
Yang is an assistant professor at Vanderbilt who leads the team that developed MutexaGPT and whose research is primarily funded by the NIGMS MIRA program.
Vanderbilt University
Vanderbilt is the Nashville-based research university where MutexaGPT was developed and where Yang joined the Arts and Science faculty.
The details
MutexaGPT elicits missing information through large language model agents and connects natural-language questions to computational tools for enzyme engineering. Enzymes are specialized proteins that speed up chemical reactions, and Yang envisions scientists engineering them as readily as consulting an AI assistant.
Timeline
Yang completed his doctorate at UCLA in four and a half years, having enrolled there in 2013.
Yang completed a postdoctoral fellowship at MIT before joining the Vanderbilt faculty in 2020.
The reference paper on the work was published in Nature Computational Science in 2026.
The Tech Race
MutexaGPT applies large language model agents, the same architecture behind general-purpose AI assistants, to enzyme engineering rather than conversational tasks. The tool is positioned to shift a field that has long depended on trial-and-error mutation testing toward computational design.
If MutexaGPT lets researchers design enzymes through plain-language requests, it could shorten the development cycle for products that rely on enzymes. The extraction does not indicate when or whether the tool will be available outside the research team.
The takeaway
Tools that connect plain-language questions to physics-based models may reduce reliance on trial and error in enzyme engineering. Readers following biotech research can watch whether such approaches move from published work into commercial use.
Further reading
Explore more Biotech coverage from Nashville for related context on research and industry developments.
