AI-Driven Autonomous Enzyme Engineering Platform Achieves 90-Fold Substrate Preference Improvement, New Study from University of Illinois Published in Nature Sub-Journal}
University of Illinois developed an AI-powered autonomous enzyme platform, boosting substrate preference by 90 times, demonstrating significant advances in biomanufacturing and synthetic biology.
Proteins are molecular machines vital to life, with broad applications in energy, health, and everyday products like detergents. However, engineering proteins for practical use remains slow, costly, and technically challenging.
Researchers at the University of Illinois Urbana-Champaign have proposed a universal autonomous enzyme engineering platform that integrates large language models (LLMs) with bio-manufacturing automation, bringing a new paradigm for rapid progress in industries like pharmaceuticals and biotech.
This research, titled A generalized platform for artificial intelligence-powered autonomous enzyme engineering, was published on July 1, 2025, in Nature Communications.
Paper link: https://www.nature.com/articles/s41467-025-61209-y
Autonomous Engineering Platform
Traditional lab efficiency depends heavily on personnel expertise, and manual experiments are labor-intensive, prone to reproducibility issues, and difficult to scale, especially with large datasets and high-throughput experiments.
AI-driven systems can explore vast multidimensional spaces more efficiently than traditional computing, while robotics and automation enable faster, more reliable experiments with higher throughput.
In biology, autonomous experiments are still in early stages, but the scalability and adaptability of universal platforms make them a priority. This study focuses on enzyme engineering to promote innovation in synthetic biology.
Using iterative Design-Build-Test-Learn (DBTL) cycles, enzymes can become more stable, selective, or efficient.
Figure 1: Overview of the autonomous protein engineering platform. (Source: the paper)
The team developed a universal platform supported by Illinois biofabrication facilities (iBioFAB), machine learning (ML), and large language models (LLMs). It only requires protein sequences and fitness data, making it applicable to a wide range of proteins.
As proof of concept, variants of Arabidopsis thaliana halogenase (AtHMT) and Yersinia pestis phytase (YmPhytase) were engineered over four weeks with four cycles, achieving 16-fold and 26-fold activity increases, respectively. Each enzyme required constructing and characterizing fewer than 500 variants.
Innovative Design and Insights
In ML-based protein engineering, sequencing and validation of point mutants often cause delays. The team designed a HiFi assembly-based mutation method that eliminates intermediate sequencing, enabling continuous workflows.
Figure 2: Automated protein engineering workflow. All steps are fully automated within the iBioFAB platform, including mutagenesis PCR, DNA assembly, and more.
Using this platform, the team demonstrated its versatility with two enzymes: AtHMT and YmPhytase. Over 50% of variants outperformed the wild type, with some showing over 16-fold and 26-fold activity increases.
Unexpectedly, ML-driven experiments often produced surprising results. For example, the best double mutant S99T/V140T was not among the top triple mutants, prompting further validation of ML’s ability to recognize synergistic effects.
Figure 3: Results of autonomous engineering of AtHMT and YmPhytase. ML predicted triple mutants with higher activity, with 82% of predictions outperforming the wild type, far exceeding the 11% of second-round mutants.
Activities increased across four cycles, with some mutants showing over 16-fold improvement for AtHMT and over 11-fold for YmPhytase. Notably, some mutants exhibited 90-fold preferences for ethyl over methyl iodide, highlighting the engineering potential.
Summary
As co-author Tianhao Yu stated, the platform features a natural language user interface, enabling users without programming skills to interact with the enzyme engineering system via simple commands, leveraging OpenAI’s API and custom functions.
This platform not only improves experimental efficiency but also demonstrates the potential for autonomous systems to revolutionize enzyme engineering.
While current limitations exist—such as applicability to all proteins—the integration of more modules in the future will establish a comprehensive, versatile framework for biomanufacturing innovation.