Enzyme Mining: The Lifeblood of Green Biomanufacturing

In the world of biotechnology, enzymes are nature's most precise 'molecular machines'—they have extremely high catalytic efficiency, strong selectivity, and work under mild conditions, making them the core driving force behind green manufacturing and synthetic biology. From stain-removing enzymes in laundry detergent to conversion enzymes in biofuels, from synthesis enzymes for pharmaceutical intermediates to eco-friendly enzymes for plastic degradation, enzymes have become indispensable 'biocatalysts' in modern industry.

However, a harsh reality is that more than 80,000 enzymes have been characterized globally, yet only about 60 are used in large-scale industrial applications—among all the enzyme molecules discovered in nature, very few are suitable for the harsh conditions of industry. How to efficiently find 'useful' enzymes in the vast natural world is the core challenge of 'enzyme mining.'

Industry data shows that the global industrial enzyme market is expected to reach around $7.66 billion in 2025 and $8.19 billion in 2026; China’s market is about 25.2 billion RMB, with growth leading the Asia-Pacific region. Behind the rapid market expansion is a 'thirsty' demand from downstream sectors like biopharmaceuticals, food processing, green chemicals, and agricultural environmental protection for new high-performance enzymes.

1. Core Definition: What is real enzyme mining?

From a professional standpoint, enzyme mining is a complete technical chain covering resource sampling → gene prediction → cloning and expression → functional validation → performance characterization → industrial adaptation. The main goal is to discover high-quality enzymes from natural biological resources that have high activity, high stability, high substrate specificity, and strong environmental tolerance, as well as to find entirely new functional enzymes in nature to fill gaps in industrial applications.

Compared to ordinary protein screening, enzyme mining is highly function-targeted: it is not limited to sequence homology comparisons but focuses on core industrial specifications such as catalytic activity, thermal stability, acid-base tolerance, substrate binding ability, and industrial suitability. It acts as a crucial bridge connecting basic biological resources with industrial applications.

Currently, in industrial enzyme applications, hydrolases account for over 60% of the industrial enzyme market, including amylases, proteases, cellulases, etc., which are mainly used for degrading macromolecular substrates in food, textiles, and feed industries. High-end specialty enzymes like oxidoreductases, transferases, and isomerases are widely used in fine chemicals, innovative drug synthesis, and high-end materials, making them a primary focus of enzyme mining today.

An unavoidable industrial reality is that by 2026, the domestic feed enzyme market has nearly 100% localization, but the import reliance for high-end specialty enzymes used in pharmaceutical synthesis and in vitro diagnostics still exceeds 70%, representing a key bottleneck in the industry.

2. Technological Iteration and Industry Pain Points: How the third-generation paradigm breaks the bottleneck

  Flowchart of the iterative process for third-generation enzyme discovery technology

After three generations of iterations, enzyme mining technology has achieved quantitative breakthroughs in R&D efficiency, screening accuracy, and resource coverage, gradually addressing the industry pain points of the traditional model of "blind mining, lengthy cycles, high costs, and scarcity of high-quality enzymes."

First Generation: Traditional Pure Training and Selection—Experience-Driven

Relying on pure microbial culture technology, cultureable microorganisms are isolated from soil, water, and extreme environmental samples, and functional enzymes are then screened through substrate color development and enzyme activity detection. This technology has a low barrier to entry and is widely used in its early stages, but it has a fatal flaw:

Under standard laboratory pure culture conditions, cultureable microorganisms account for less than 1%—this is the well-known "Great Plate Count Anomaly" in microbiology. Vast amounts of high-quality enzyme resources are completely overlooked, with screening cycles of 1-3 years, extremely high trial-and-error costs, and cannot meet the rapid iteration needs of industrialization.

A typical example of this period was the screening of microorganisms in extreme environments: searching for thermophiles and halophiles from volcanic craters, deep-sea hydrothermal vents, and salt lakes. Although important industrial enzymes such as Taq DNA polymerase and thermophilic lipase were produced, the throughput was extremely low and the cycle was extremely long, with a single project often taking several years.

Second Generation: Metagenomic Bioinformatics Mining—Initial Data Drive

With the widespread adoption of genome sequencing technology, the industry has overcome the limitations of pure cultivation, obtaining all genetic information from environmental samples through metagenomic sequencing. Relying on bioinformatics methods such as sequence homology alignment, conserved domain analysis, and functional annotation, it predicts potential functional enzymes. Resource coverage has greatly increased, raising the resource pool for enzyme mining from "tens of thousands of strains" to "hundreds of millions of genes."

There are two specific paths: sequence-driven screening has high throughput and speed, but can only detect "known analogs" and is difficult to discover entirely new families; Function-driven screening can discover enzymes with entirely new functions, but the throughput is extremely low, the workload is high, and the expression system is limited.

The core issue is the disconnect between sequence prediction and actual function: a large number of homologous sequences lack true catalytic activity, and traditional homologous comparison methods like BLAST make it difficult to identify distant or entirely new families of functional enzymes. In the microbial genome and metagenome, the functions of the vast majority of proteins have yet to be experimentally characterized—this portion is called "microbial dark matter," which contains vast amounts of unknown enzyme resources.

Third Generation: AI Intelligent Targeted Enzyme Mining—Precise Implementation

The most advanced industrialization technology paradigm today relies on massive protein sequence databases, enzyme function annotation datasets, and deep learning models to build a full-dimensional mapping from sequence to structure to function to industrial performance. This breaks free from the limitations of traditional homology comparisons and enables the discovery of entirely new enzyme resources across families without templates.

AI technologies represented by protein language models (PLMs) can be pretrained on billions of protein sequences to learn deep relationships between sequences and functions. For enzyme discovery, this means three major breakthroughs:

- Cross-family function prediction: Even when the target sequence has low similarity to known enzymes (in some cases <30%), the model can still predict catalytic capabilities based on global sequence features, evolutionary co-variation signals, and potential functional site information, greatly expanding the search boundaries of traditional homology comparison.

- Multi-attribute joint evaluation: Not just whether there’s activity, but also optimal temperature, pH, substrate spectrum, stability, and other industrial properties.

- Seamless transition from 'discovery' to 'modification': Once candidate enzymes are found, the same model system can be used for directed evolution optimization, integrating discovery, modification, and validation as a whole.

Single-sequence function prediction can be done in milliseconds, and screening a library of millions of sequences can be completed within hours. Combined with automated experimental platforms, the full validation cycle that traditionally takes 1-3 years can be shortened by over 60%, while precisely targeting industrial-grade enzymes that are heat-resistant, acid-alkali resistant, and highly catalytically specific. This truly achieves 'on-demand discovery, precise matching, and efficient implementation.'

Industry consensus is that AI in enzyme engineering has formed a technical route of 'pathfinding → enzyme discovery → enzyme modification → enzyme creation,' with enzyme discovery and modification currently being the most commercially mature steps.

3. From 'Discovery to Use': The Industrialization Chain of Enzyme Mining

Finding an active enzyme is just the first step of a long journey. From an enzyme with activity in the lab to one that works on an industrial production line, there’s a whole transformation chain in between, and each step is a hurdle.

The first hurdle: heterologous expression. Genes dug out from the metagenome often come from unculturable microbes. When transferred to common expression hosts like E. coli or Pichia pastoris, problems like no expression, low expression levels, or inclusion body formation may occur. Codon optimization, host screening, co-expression with molecular chaperones… each optimization requires a lot of trial-and-error experiments.

The second hurdle: stability and tolerance. Enzymes that are active in the lab might quickly become inactive in industrial reactors. High temperature, extreme pH, high substrate concentration, organic solvents… industrial environments are far harsher than labs. A qualified industrial enzyme needs to meet multiple criteria at the same time: high specific activity, high stability, high expression levels, low production cost, controllable substrate specificity, etc. Multi-objective optimization is an eternal challenge in enzyme engineering.

The third hurdle: large-scale production and cost. The fermentation yield, purification cost, and recycling efficiency of enzymes directly determine whether they are affordable for industrial use. Take immobilized enzyme technology as an example: a good immobilization scheme can both enhance enzyme stability and enable enzyme reuse, cutting overall costs by more than half.

This is also why the industry consensus is becoming clearer: a single technological breakthrough is not enough; a systematic solution integrating data, algorithms, and experiments is needed.

4. Platform enablement: MatwingsVenus™ (Xiaowu™) builds an end-to-end AI enzyme mining solution.

 End-to-end AI enzyme discovery platform

So, is it possible to integrate massive data, AI models, and automated experiments into a single platform, turning enzyme discovery from a 'handcrafted workshop' into a 'production line'? Shanghai Matwings Technology's MatwingsVenus™ (Xiaowu™) offers a domestic answer.

Its enzyme-mining capabilities are built on three layers of infrastructure.

The first layer: the ultra-large protein dataset, MatwingsVenus™ Pod. According to Shanghai Jiao Tong University's announcement of the '5th Top Ten Scientific and Technological Advances,' this platform gathers exclusive extreme environment data, including the deep-sea MEER project and hypersaline or polar microorganisms. It has cleaned and integrated 15 billion protein sequences worldwide, including 6.5 billion high-quality sequences labeled with environmental information like temperature and pH.

What does having labeled data mean? Traditional enzyme mining can only answer 'Is this sequence an enzyme?' Labeled data can further answer 'Under what conditions does this enzyme work?' — which is crucial for industrial enzyme screening.

The second layer: the AI enzyme-mining model, MatwingsVenus™ Large Model. Built on the self-developed VenusPLM general protein model (from Prof. Hong Liang’s team at Shanghai Jiao Tong University), the platform has developed specialized enzyme mining capabilities. It can efficiently mine enzymes with low sequence similarity but great function from massive protein databases, and it can accurately predict protein stability, activity, and expression even with zero or few examples. Compared to methods that rely solely on sequence similarity, the coverage and accuracy of functional predictions are significantly improved.

Candidate enzymes that don't meet performance standards can enter the AI-guided evolution optimization process, optimizing activity, stability, and expression simultaneously, enabling a seamless 'discover-and-improve enzyme' workflow.

The third layer: the wet-dry closed-loop validation system, MatwingsVenus™ Auto. More importantly, the platform doesn't just provide computational predictions—it connects to automated experimental validation. Released in April 2026, the MatwingsVenus™ (Xiaowu™) agent enables a 'design equals validation, validation equals iteration' R&D mode: users state their needs in natural language, the agent calls the enzyme-mining model to filter candidate sequences, then drives an automated wet-lab platform to complete expression, purification, and functional tests. Results flow back to the next AI design iteration, forming a complete 'conversational wet-dry closed loop.'

The core value of this model is that it compresses the full 'discover-improve-test enzyme' process, which used to be achievable only by large corporate R&D teams, into a process accessible even to individuals. Public information shows that the platform has successfully delivered over 30 high-performance proteins, with more than 10 achieving large-scale production.

Specifically in the enzyme mining field, there are already several verifiable real-world cases: In the PET plastic-degrading enzyme project, AI identified a novel PET hydrolase from massive sequence data, which after directed evolution saw a 97-fold increase in enzyme activity, with an R&D cycle of just 6 months; in the food field’s glycosyltransferase engineering project, the platform boosted the enzyme’s overall glycosylation activity 7-fold in just 4 months, improved product specificity from 60% to 98%, and reduced core material costs by 90%—whereas traditional methods usually take 2-3 years.

5. Industry Value: AI Enzyme Mining Enables Green Upgrades Across All Sectors

High-quality enzyme resources are a core necessity for the bioindustry. AI-driven efficient enzyme mining technology is enabling technological innovation across multiple sectors, continuously releasing industrial value.

Industrial Biomanufacturing: Mining highly active, highly tolerant industrial enzymes to replace traditional chemical catalysts, significantly reducing energy consumption and carbon emissions in chemical synthesis, biomass conversion, and biodegradation processes, while improving product conversion rates and purity. Enzyme-catalyzed reaction rates are 10⁶–10²⁰ times faster than non-catalyzed reactions (depending on the enzyme type), 10⁷–10¹³ times higher than traditional chemical catalysts, with heavy metal and hazardous waste emissions reduced by over 90%, supporting the green transformation of traditional industries.

Biopharmaceuticals: Precisely mining specific synthetic enzymes and modification enzymes to fit the synthesis processes of small molecule drugs, peptide drugs, and antibody-drug conjugates, improving drug synthesis efficiency and reducing impurity production. Given the current situation where over 70% of high-end specialty enzymes for pharmaceutical synthesis are imported, domestic AI enzyme mining platforms are becoming key to breaking this dependence, aiding in the low-cost, high-quality production of innovative drugs.

Food and Daily Chemicals: Mining mild, efficient, and safe food-grade enzymes for use in food modification, preservation, deep processing, and daily chemical cleaning applications. Amylases and saccharifying enzymes are widely used in starch syrup, brewing, and baking processes, improving raw material conversion rates by 10%-20%; enzyme-processing replacing chemical additives not only enhances product quality but also meets consumers’ demand for clean-label products.

Environmental Management: Mining degradation enzymes adapted to extreme environments to efficiently biodegrade plastics, oil, and organic pollutants. Take PET-degrading enzymes as an example: AI-assisted directed evolution can significantly improve their thermal stability and catalytic efficiency, offering green technology solutions for solid waste recycling and ecological restoration.

6. Industry Outlook: Enzyme Mining, the Original Innovation Engine of the Bioeconomy

Currently, global competition in the bioindustry has focused on innovation of source resources. As core biological catalysts, enzymes' mining efficiency and innovation capability directly determine the core competitiveness of the biomanufacturing industry. The traditional approach of relying on natural screening and importing overseas enzymes can no longer meet the domestic industry's need for independent control and high-end upgrades.

In the future, AI-driven intelligent enzyme mining will become the mainstream paradigm, shifting from 'passively screening natural enzymes' to a new stage of 'actively discovering, precisely optimizing, and custom-creating' enzymes. Domestic independent AI enzyme mining platforms like MatwingsVenus™ (Xiaowu™) will continue to leverage data and algorithm advantages to solve industry challenges such as the scarcity of high-quality industrial enzymes in China, long R&D cycles, and high implementation costs, providing research institutions, biopharmaceutical, and biomanufacturing companies with efficient, low-cost, and practical end-to-end enzyme mining solutions.

Mining tiny catalytic forces from massive biological genes, high-quality enzyme resources empower the trillion-level bioindustry. This upstream technological revolution in enzyme mining will continue to stimulate the innovation vitality of green biomanufacturing and drive China's bioindustry to leap from technology follower to source leader.