Say goodbye to blind trial and error! How can AI accurately activate glycotransferase activity?

In key industry scenarios like large-scale production of functional oligosaccharides, glycosylation modification of natural active molecules, and green enzymatic synthesis of high-end human milk oligosaccharides, transglycosylation activity has clearly become a core technology that supports the global functional sugar industry's iterative upgrades. The practical application of AI-based protein design technology is pushing the field away from the long-standing trial-and-error R&D approach toward precise, customized solutions.

1. Decoding Transglycosylation Activity: A Molecular Tool That Can Both 'Hydrolyze' and 'Transfer'

Glycosidase Two Reaction Pathways

Transglycosidases are an important functional subtype of glycoside hydrolase (GH) families, with a dual catalytic feature: they can catalyze the hydrolysis of glycosidic bonds and, under suitable conditions, transfer sugar units to various acceptor molecules to form new glycosidic bonds. They differ fundamentally from glycosyltransferases (GTs, EC 2.4.x.y), which are dedicated to synthesizing glycosidic bonds: glycosyltransferases require activated sugar nucleotides like UDP-glucose to complete the reaction, which is costly; transglycosidases, on the other hand, can directly use ordinary sugars as donors, offering milder reaction conditions and lower raw material costs.

Although transglycosidases avoid the cost dependency of activated sugar nucleotides, their catalytic mechanism naturally comes with a challenging engineering issue—the core evaluation metric is transglycosylation selectivity. This measures the enzyme's ability to favor synthetic transglycosylation reactions while suppressing ineffective hydrolysis, directly determining the industrial value of the enzyme. The two types of reactions are clearly defined:

- Hydrolysis reaction: the sugar unit from the donor is transferred to water, only breaking down the substrate, which is an unproductive side reaction;

- Transglycosylation reaction: the sugar unit from the donor is directed to the acceptor molecule, extending sugar chains or modifying molecules to produce the target product.

The higher the transglycosylation selectivity, the lower the proportion of unproductive hydrolysis, the higher the effective synthesis efficiency, and the stronger the industrial adaptability of the enzyme.

In terms of activity detection, the quantitative system for transglycosylation activity is adapted depending on the enzyme type and application scenario. Mainstream industry detection methods include high-performance liquid chromatography (HPLC) and liquid chromatography-mass spectrometry (LC/MS), with some research applications using fluorescent substrates for assisted detection. The core logic of all detection methods is consistent: precisely quantify the total amount of effective sugar products generated by the enzyme per unit time, thereby evaluating the enzyme's catalytic activity and overall application performance.

2. Traditional R&D Dilemma: Blind trial-and-error limits industrial implementation

Naturally sourced transglucosidases generally have industrial adaptation issues, showing poor thermal stability, imbalanced selectivity between transglycosylation and hydrolysis reactions, limited substrate compatibility, and difficulty recognizing non-natural substrates, making them unable to directly meet the needs of large-scale, low-cost, high-purity industrial production.

Before the implementation of AI technology, the industry mainly relied on two traditional methods to modify transglucosidases, but both faced hard-to-overcome efficiency bottlenecks:

Directed evolution relies on random gene mutations to build massive mutant libraries and screens for high-quality variants through extensive wet experiments. However, the amino acid sequence space of transglucosidases is extremely large; single-site and multi-site mutations can generate billions of candidate samples, posing high trial-and-error costs, long screening cycles, and extremely low chances of finding optimal variants. A complete iteration often takes 2 to 3 years.

Rational design depends on the static 3D structure of proteins for targeted amino acid modifications, but its core limitation is ignoring the dynamic nature of the catalytic system—the catalysis by transglucosidases is a dynamic process involving the coordinated action of multiple residues, with protein conformations continuously adjusting during the reaction. Modifying a single target site can easily disrupt overall catalytic balance, often resulting in a slight improvement in one performance aspect while causing a significant drop in overall performance.

In summary, traditional enzyme modification techniques cannot escape the inherent limitations of 'random trial-and-error and passive screening,' making it difficult to optimize multiple performance indicators simultaneously and long-term restricting the industrial application of high-performance transglucosidases.

3. Core Industrial Value: Transglucosidases empower four major high-growth application scenarios.

Gradually Brightening Glycan Chain

The core value of glycosyltransferases lies in their unique ability to gently and precisely edit sugar chains. Traditional chemical synthesis often faces issues like cumbersome reaction steps, poor regioselectivity, many by-products, and high energy consumption. Glycosyltransferases, on the other hand, can complete directional sugar chain extension and precise modification in a one-step reaction under normal temperature and pressure in aqueous environments, providing green and efficient biocatalytic solutions for various fields:

3.1. Enzymatic Synthesis of Functional Oligosaccharides

In the enzymatic synthesis of functional oligosaccharides, different types rely on different glycosyltransferase pathways: the production of FOS depends on fructosyltransferase (GH32/GH68) using sucrose as the donor to perform the fructosyl transfer reaction; the production of GOS relies on β-galactosidase (GH1/GH2) using lactose as the donor for galactosyl transfer; cyclodextrin glucanotransferase (CGTase, EC 2.4.1.19, GH13) is commonly used for the enzymatic synthesis of cyclodextrins and some malt oligosaccharides. This enzyme belongs to the GH13 α-amylase family in the CAZy database. Compared to traditional chemical synthesis, enzymatic catalysis can achieve directional sugar chain extension in a single step, greatly simplifying the process, reducing energy consumption, and cutting production costs, making it the mainstream green production solution recognized by the industry today.

3.2. Biomanufacturing of Human Milk Oligosaccharides (HMOs)

Human milk oligosaccharides are the key active nutritional components in breast milk and a crucial raw material in upgrading infant formula to closer match breast milk. It's one of the fastest-growing niches in the functional sugars sector. Industrial production of HMOs currently mainly relies on GT family glycosyltransferases as catalysts, using engineered microbes’ sugar nucleotide regeneration systems for efficient synthesis. GH family glycosyltransferases are being explored as a complementary strategy in certain reactions, such as reverse glycosylation of fucosylated oligosaccharides, which may further reduce raw material costs and provide core support for large-scale, low-cost HMO production.

3.3. Glycosylation Modification of Natural Active Compounds

Natural active substances like ginsenosides, steviol glycosides, and flavonoids often suffer from poor water solubility, low bioavailability, and unpleasant taste. Using glycosyltransferases for precise glycosylation modifications can optimize water solubility, taste, and stability while retaining bioactivity, significantly increasing their application value. Enzymatic modification of steviol glycosides is a mature benchmark case in the industry, where directional glycosylation eliminates bitter aftertaste, promoting the widespread household use of natural zero-calorie sweeteners.

4. Paradigm Shift: AI Enables “On-Demand Design” for Precision R&D

The deep implementation of large AI protein models has systematically reconstructed the entire R&D system for transglycosylases, transforming the traditional inefficient mode of 'random mutation and passive screening' into an intelligent model of 'goal-oriented, on-demand customization':

Capability 1: Rapid structure analysis and precise mechanism targeting

AI structure prediction tools, represented by AlphaFold, can accurately determine the complete 3D structure of transglycosylases in a short time, pinpointing key functional areas like substrate-binding pockets and core catalytic residues. Compared to traditional crystal diffraction that can take months, AI can quickly complete structural modeling and mechanism analysis, providing theoretical support for targeted modifications.

Capability 2: Reverse function customization and optimal sequence deduction

Unlike the traditional 'sequence-structure-function' forward analytical logic, AI protein design models can efficiently predict mutation effects across a large candidate sequence space, selecting optimized sequence combinations that meet multiple constraints, such as high transglycosylase activity, low hydrolysis side reactions, and high substrate specificity, significantly reducing experimental trial-and-error costs.

Capability 3: Multi-metric collaborative optimization to solve trade-offs

Industrial transglycosylases need to simultaneously meet multiple constraints such as high activity, high selectivity, and high stability, which naturally have trade-offs. AI can globally solve across ultra-large sequence spaces, achieving collaborative upgrades of multiple performance metrics, adapting to complex industrial production scenarios.

5. Industrial evidence: AI-driven leap in transglycosylase performance

The value of AI protein design technology in optimizing transglycosylases has been validated by multiple industrial projects and research results:

Case 1: Tianwu Technology — Multi-Metric Synergistic Optimization and Industrial Implementation

According to a case published on the official website of Tianwu Technology, the research team relied on an AI protein design platform to conduct targeted iterative optimization on industrial-grade glycosyltransferases. In just four months, they achieved a 7-fold increase in overall transglycosylation activity, raised product specificity to 95%, and reduced hydrolytic activity by 33%. Compared to the traditional 2–3 year modification cycle, AI development cut the timeline by over 80%, while also solving the industry challenge that 'increasing transglycosylation activity inevitably comes with an increase in hydrolytic side reactions.'

Case 2: Pro-PRIME — Protein Language Model Solves Reaction Balance Challenge

Research published in Bioresource Technology (Volume 438, December 2025, 133206) showed that researchers screened a cyclodextrinase (CDase, EC 3.2.1.54) from Paenibacillus sp. MY03 with naturally high transglycosylation/hydrolysis ratios and used the Pro-PRIME protein language model for precise global modifications. This optimized transglycosylation activity, hydrolysis side reaction control, and regioselectivity all at once. Using only a small amount of beneficial mutation data, AI simultaneously enhanced these three core performance metrics. The best mutant achieved a 12-fold increase in the transglycosylation/hydrolysis (T/H) ratio, and the pNP-G7 yield rose from 63% to 98%, systematically addressing multiple performance trade-off issues.

6.Industry Outlook: AI Opens a New Era of Programmable Glycan Manufacturing.

Sugar Chain Multi-Node Editing Workflow

Currently, the development of transglycosylases is at a pivotal point, shifting from the traditional experience-driven model to an AI-driven precise and programmable design model:

In the functional food field: AI-customized transglycosylases will continue to lower the production costs of functional oligosaccharides and modified sweeteners, driving the expansion and upgrading of industries like prebiotics and natural sweeteners;

In the biopharmaceutical field: high-precision transglycosylases can enable accurate glycosylation of drugs and antibody drugs, improving drug safety and effectiveness;

In the green chemical field: leveraging the mild catalytic advantage of transglycosylases, renewable sugar-based raw materials can be used to synthesize complex glycosidic compounds, replacing highly polluting traditional processes.

In terms of R&D efficiency, developing a mature industrial-grade transglycosylase using traditional techniques takes 2 to 3 years, while an AI-driven development system can compress the cycle to just a few months, significantly speeding up the iteration of new functional sugar products.

Conclusion

The essence of transglycosylase activity lies in humanity's core ability to actively edit and reshape the functions of biological sugar chains. It is the key technological cornerstone of the functional sugar industry and glycoengineering. From passive screening of natural enzymes, to artificial modification, and then to active precise creation with AI, transglycosylase development has achieved a significant leap. AI protein design technology is effectively breaking through industry bottlenecks that involve balancing multiple performance factors, turning traditional enzymology into a quantifiable, designable, and iterative modern engineering science. As high-performance AI-modified transglycosylases gradually enter industrial fermentation workshops, transglycosylase activity has shifted from a natural intrinsic property to a core technological capability that can be programmed and directionally optimized—this is exactly the driving force behind the high-quality development of the functional sugar industry.