Project Description

I. Background and Problem

 Excessive intake of too many high-sugar, high-fat foods has made obesity, hyperlipidemia, hypertension, and diabetes much more common around the world. A key way to improve public health is to cut down on calories by using low-calorie functional sugars. D-allulose (D-psicose) is one of more than 30 known rare sugars. It is unique because it is 70% as sweet as sucrose but has almost no calories. It also has good effects on the body, like controlling lipid metabolism, protecting the brain, and fighting obesity. As a result, D-allulose is now a safe food additive in many countries. In July 2025, it has officially been recognized in China as a new food ingredient. This is a big step forward for synthetic biology in the food industry.

 However, the industrial-scale production of D-allulose remains limited by the catalytic efficiency and stability of its key enzyme, D-tagatose-3-epimerase (DTE), which catalyzes the epimerization of D-fructose into D-allulose. Conventional directed evolution strategies for enzyme improvement are time-consuming, labor-intensive, and often yield limited performance gains. To enable large-scale, efficient bioproduction of D-allulose, a new approach combining computational enzyme design and experimental validation is urgently needed.

II. Project Goals and Objectives

 To overcome the limitations of traditional enzyme engineering, our project integrates artificial intelligence (AI) with experimental validation to achieve de novo design and functional optimization of DTE. Our main goals are:

• AI-driven enzyme design using LigandMPNN for sequence redesign, AlphaFold3 for structural prediction, molecular docking for substrate binding assessment, and molecular dynamics (MD) simulations for stability evaluation.
• Experimental validation of designed DTE variants to confirm their catalytic activity and substrate conversion efficiency.
• Development of a Pepper fluorescent RNA-based biosensor to monitor and quantify enzymatic reactions in real time.
• Construction of computational models, including the Catalytic Circuit Model, Characterization Circuit Model, Population Dynamics Model, and Half-life Integration Model, to describe and optimize the entire system.

 Through these steps, our team aims to establish a complete AI-to-wet-lab workflow for rapid enzyme optimization and rare sugar biomanufacturing.

III. Inspiration and Project Rationale

 Our project was inspired by the increasing global demand for healthier sugar alternatives and the recent milestone approval of D-allulose as a new food ingredient in China. Despite its remarkable nutritional and physiological benefits, the industrial production of D-allulose remains constrained by the limited catalytic performance of DTE. Conventional approaches, such as random mutagenesis and directed evolution, are often slow and inefficient, making them unsuitable for rapid enzyme optimization.

 Recognizing the rapid advances in deep learning and AI model, our team decided to apply cutting-edge artificial intelligence tools and protein modeling technologies to redesign DTE. By using LigandMPNN, AlphaFold3, molecular docking, and molecular dynamics simulations, we aimed to explore novel enzyme variants with improved structural stability and catalytic potential.

 This approach represents a new application of AI-assisted enzyme engineering in the context of rare sugar biosynthesis. It demonstrates how advanced protein structure prediction tools and de novo design models can be systematically integrated with experimental validation to accelerate the design-build-test-learn (DBTL) cycle. Through this project, we hope to build a practical, reproducible workflow for future synthetic biology teams to combine AI-based analysis with wet lab validation.

IV. Scientific and Technical Approach

 The project follows the DBTL (Design-Build-Test-Learn) engineering cycle recommended by iGEM, enabling continuous refinement of our system.

 Design: The dry lab team applied LigandMPNN for AI-based sequence redesign of DTE, focusing on active-site residues to improve substrate affinity and catalytic efficiency. The predicted structures were generated using AlphaFold3, evaluated via molecular docking, and further tested through molecular dynamics simulations to assess conformational stability and flexibility.

 Build: The wet lab constructed plasmids for expressing the designed DTE variants and developed a Pepper fluorescent RNA-based fructose biosensor capable of detecting intermediates produced during the enzymatic reaction.

 Test: By integrating the biosensor and enzyme systems, we quantitatively measured the catalytic activity of AI-designed DTEs and validated their performance in vitro.

 Learn: Experimental data provided feedback for refining both the computational models and AI design parameters, closing the loop for the next round of optimization.

V. Modeling and System Integration

 To describe and predict our system, we developed four mathematical models, including Catalytic Circuit Model, Characterization Circuit Model, Population Dynamics Model, and Half-life System Integration Model. These models enable data-driven optimization of both enzyme performance and biosensor design, ensuring robust system integration between computational and experimental domains.

VI. Potential Impact

 By combining artificial intelligence with experimental biology, our project shows a practical path toward faster and smarter enzyme design. Through the redesign and validation of D-tagatose-3-epimerase, we not only improved the catalytic potential for D-allulose production but also built a framework that other teams can easily adapt for enzyme engineering. This work bridges computation and wet lab experimentation, helping make biomanufacturing more efficient, sustainable, and creative. Beyond D-allulose, we believe this approach can inspire new ways to apply AI tools in synthetic biology, turning complex data into tangible solutions for food, health, and the environment.