Catalytic Activity: The Core Measure of How Well an Enzyme Works

Enzymes are microscopic molecular machines that drive the biochemical operations of living systems. By virtue of their remarkably efficient catalytic properties, they ensure that various chemical reactions proceed rapidly and orderly within the body. Enzyme catalytic activity serves as the core quantitative metric for assessing the working efficiency of enzyme molecules, and it is an indispensable evaluation criterion across three major domains: industrial enzyme engineering, green biomanufacturing, and in vitro clinical diagnostics.

Traditional enzyme engineering has long been constrained by inherent bottlenecks—lengthy screening cycles, high R&D costs, and the stochastic nature of mutation directions. The deployment and iterative advancement of AI-powered protein design technologies have effectively broken through these traditional constraints, enabling precise directed optimization and programmable iterative upgrading of enzyme catalytic activity, thereby laying a solid technological foundation for the large-scale industrialization of industrial enzymes.

I. Core Quantitative Metrics for Enzyme Catalytic Activity: A Standardized System for Evaluating Industrial Enzyme Performance

Enzyme Activity Dual-State Comparison

Enzyme catalytic activity is not a single numerical value but rather a complete, standardized system of kinetic quantitative metrics that spans the entire chain from fundamental enzymology research to industrial-scale production and in vitro clinical diagnostics. It serves as the core bridge connecting laboratory basic research to commercial industrial application. From an industrial perspective, catalytic activity directly determines the substrate conversion throughput per unit time, serving as the core parameter for differentiating enzyme product quality and controlling raw material and energy costs.

In the clinical diagnostics track: The catalytic activity concentration of serum alanine aminotransferase (ALT) is a core diagnostic marker for clinically assessing the extent of hepatocyte injury.

In the food fermentation, biopharmaceutical, and biochemical engineering tracks: The catalytic activity of industrial enzymes—including amylases, proteases, lipases, and glycosyltransferases—directly determines the production capacity ceiling and overall production costs of manufacturing lines, constituting the core competitive barrier for enterprise enzyme products.

Three standardized kinetic metrics are universally adopted across the industry, collectively defining the boundaries of enzyme catalytic capability from the dimensions of single-molecule efficiency, substrate-binding affinity, and industrially measured performance:

Turnover Number (kcat) — Core Metric for Single-Molecule Catalytic Efficiency

Defined as the maximum number of substrate molecules converted to product per active site per second under substrate-saturated conditions, with the unit s⁻¹. A higher kcat value indicates greater intrinsic catalytic efficiency of the enzyme molecule. Carbonic anhydrase is one of the most catalytically efficient natural enzymes known, with each molecule converting approximately 10⁶ substrate molecules per second, corresponding to kcat ≈ 1 × 10⁶ s⁻¹.

Michaelis Constant (Km) — Criterion for Enzyme–Substrate Binding Affinity

Km represents the substrate concentration at which the reaction velocity reaches half of Vmax. In typical enzyme-catalyzed reactions where the catalytic step is rate-limiting (k₃ < k₂), Km approximates the dissociation constant Kd of the enzyme–substrate (ES) complex. In this context, a lower numerical value indicates stronger enzyme–substrate binding affinity and a lower threshold for catalytic initiation.

Enzyme Activity Units (IU/kat) — Industrially Intuitive Practical Measurement Standard

International Unit (IU): Under optimal standardized reaction conditions, the amount of enzyme required to catalyze the formation of 1 μmol of product per minute is defined as 1 IU.

SI unit (katal, kat): The amount of enzyme required to catalyze the conversion of 1 mol of substrate per second, with the conversion relationship: 1 kat = 6 × 10⁷ IU.

Critical Prerequisite: Enzyme catalytic activity is highly sensitive to environmental conditions—temperature, pH, substrate concentration, and buffer ionic strength can all significantly alter measured activity values. All cross-comparisons of enzyme activity and kinetic parameter determinations must be conducted under uniformly standardized reaction systems for the data to possess comparative reference value.

II. Bottlenecks in Traditional Enzyme Activity Engineering: High Trial-and-Error Costs and Protracted Iteration Cycles

Peak Catalytic Enzyme Luminescence

Naturally occurring wild-type enzymes generally suffer from limitations, including insufficient thermostability, narrow substrate specificity, suboptimal catalytic efficiency, and poor acid–base tolerance, making them difficult to directly adapt to the demanding conditions of continuous industrial production. Enhancing enzyme catalytic activity has long been the core R&D objective of modern enzyme engineering. However, both traditional engineering approaches face inherent efficiency limitations that are difficult to overcome:

1. Directed Evolution: Random Mutagenesis Libraries, Needle-in-a-Haystack Screening

This approach employs random gene mutagenesis to construct massive mutant libraries, followed by high-throughput experimental screening to identify rare high-activity variants. However, the protein sequence space is virtually infinite—random mutagenesis screening is tantamount to blind searching for a target variant within a vast sequence landscape. A complete round of engineering plus screening typically spans several years. For example, in the engineering of sulfotransferases, conventional homologous sequence BLAST screening of 1,000 homologous natural enzymes yields extremely low hit rates for high-activity variants, and the R&D pace lags far behind the rapid iteration demands of the industry.

2. Rational Site-Directed Mutagenesis: Single-Target Engineering Prone to Performance Imbalance

This approach relies on resolved three-dimensional structures to target individual amino acids at the catalytic site for site-directed mutagenesis. However, enzyme catalysis is a dynamic, synergistic process involving multiple residues within the active center. Single-point mutations frequently result in performance imbalances—such as "increased activity accompanied by plummeting stability" or "elevated catalytic efficiency with loss of substrate selectivity." The success rate of such engineering attempts remains low, and repeated experimental cycles drive up R&D costs.

Collectively, both directed evolution and rational design are heavily dependent on human trial-and-error and the empirical judgment of experienced researchers—characterized by long timelines, high material costs, and low success rates in isolating beneficial mutants—which have long constrained the iterative upgrading of industrial enzyme catalytic performance. AI-driven next-generation protein design technologies are now systematically reshaping the industry's R&D paradigm.

III. AI-Driven Enzyme Molecular Design: Catalytic Activity Transitioning from Empirical Trial-and-Error to Precision Programmability

The continuous iteration of generative AI protein models is propelling enzyme catalytic activity engineering beyond traditional manual screening paradigms into a new phase of precision-on-demand design, integrating three dimensions: mechanistic elucidation — zero-shot de novo design — multi-property synergistic optimization. Leveraging three foundational technologies—deep learning, protein structure prediction, and diffusion generative models—AI effectively overcomes the pain points of conventional engineering approaches and enables multi-dimensional targeted enhancement of enzyme catalytic activity.

1. AI Atomic-Level Structural Resolution, Precisely Deconstructing Catalytic Mechanisms

In February 2025, Science published a breakthrough study in which researchers combined the PLACER machine learning network with the RFdiffusion diffusion model to achieve de novo design of serine hydrolases with complete catalytic function. PLACER (Protein–Ligand Atomistic Conformational Ensemble Resolver) can simultaneously analyze protein backbone physicochemical properties, amino acid side-chain characteristics, and substrate molecular chemical configurations, precisely predicting the three-dimensional spatial arrangement and coordination geometry of enzyme active-site atoms. Under low-throughput screening conditions, the study uncovered five enzyme folding conformations previously unknown in nature, substantially expanding the structural repository available for high-activity industrial enzyme design.

Building upon the serine hydrolase work, the next-generation RFdiffusion2 diffusion model further enabled atomic-level customized design of active sites. A December 2025 study in Nature Methods validated this approach for zinc metallohydrolase design: from only 96 AI-designed samples in the first round, the high-activity variant ZETA_1 was obtained, exhibiting a catalytic efficiency (kcat/Km) of 16,000 M⁻¹s⁻¹. A second iteration generated three performance-enhanced mutants, with the best variant ZETA_2 achieving a catalytic efficiency of 53,000 M⁻¹s⁻¹. In just two rounds of AI design, the ZETA_2 variant—with a kcat/Km of 5.3 × 10⁴ M⁻¹s⁻¹—set a new activity record for de novo designed metallohydrolases, representing an order-of-magnitude leap over the typical sub10²–10³ M⁻¹s⁻¹ levels previously achieved by AI-designed enzymes.

2. Zero-Shot De Novo AI Design: Rapid Generation of High-Activity Novel Enzyme Sequences Without Natural Templates

Traditional engineering relies on natural homologous enzyme templates and requires multiple rounds of mutagenesis, screening, and validation cycles. Advanced generative AI models can directly generate entirely new high-activity enzyme sequences based solely on the input of catalytic function objectives, enabling zero-shot high-efficiency development—without the need for natural protein scaffolds or pre-specified homologous templates, directly outputting candidate sequences with atomically precise catalytic sites. The aforementioned RFdiffusion2 metallohydrolase case is a typical example: two rounds of AI design achieved catalytic efficiency levels that would have required years of traditional engineering, compressing the R&D timeline by over 90%.

3. Multi-Objective Synergistic Global Optimization, Adapting to Complex Industrial Conditions

Industrial enzymes require not only high catalytic activity but must also simultaneously satisfy multiple performance criteria—substrate selectivity, thermostability, acid–base tolerance, and resistance to product inhibition. Enhancing a single dimension of activity often compromises other properties. AI models enable multi-objective parameter co-optimization, avoiding the trade-offs inherent in single-property engineering.

A 2026 study in Chemical Engineering Journal addressed the low hit-rate problem of traditional screening: deep learning models can directly and accurately predict comprehensive enzyme kinetic parameters, enabling pre-screening of superior candidate sequences. Under this framework, the hit rate for high-activity enzyme variants reached 50%—an order-of-magnitude improvement over the <1% hit rate of traditional random mutagenesis. Subsequent minor directed evolution iterations further enhanced the target enzyme's catalytic activity up to 7.25-fold, with engineered E. coli strains achieving high-yield production of 1578 μM sulfotyrosine—demonstrating a full-chain technical enablement from "AI molecular design → strain construction → industrial fermentation."

IV. Industrial Commercialization Cases: Large-Scale AI Empowerment of Industrial Enzyme Activity Upgrading

Dry-Wet Iteration Loop for Enzyme Engineering

AI-directed enzyme catalytic activity optimization technologies have now moved beyond the laboratory and achieved commercial deployment across three major tracks—industrial enzyme preparations, innovative biopharmaceuticals, and in vitro diagnostic reagents—with leading enterprises and cutting-edge research outcomes fully validating the industrial value of AI-enabled enzyme design.

1. Matwings Technology's MatwingsVenus™ （晓鹜™） Conversational Protein R&D Platform — Application Cases

According to publicly disclosed project data from the company, the proprietary MatwingsVenus™ AI protein R&D platform has achieved multiple industrial breakthroughs in multi-dimensional industrial enzyme performance optimization:

Glycosyltransferase engineering project: Within only 4 months of R&D, total transglycosylation catalytic activity was enhanced 7-fold, target product specificity reached 95%, and side-reaction hydrolytic activity was reduced by 33%, fully meeting the continuous production standards of the food and biopharmaceutical industries.

Functional antibody molecular optimization project: Over 4 months, a single-domain antibody (nanobody) achieved a 400% improvement in alkaline tolerance and a 1.2-fold increase in antigen-binding affinity.

To date, Matwings Technology has successfully delivered over 30 complete protein molecular design projects, with nearly 10 proprietary enzyme products achieving commercial production. These cover core tracks including innovative drug target proteins, IVD diagnostic proteins, and commodity industrial enzymes. Multiple proprietary enzyme products have achieved catalytic activity metrics that meet or exceed those of comparable commercial products from leading international companies.

2. Global Cutting-Edge Academic Achievements: AI Continually Pushing the Ceiling of Industrial Enzyme Catalytic Efficiency

The global research community continues to push the performance ceiling of AI-engineered enzymes. A 2026 study published in Bioresource Technology demonstrated that a generative AI-engineered β-thioglucosidase variant exhibited over a 9-fold improvement in catalytic efficiency. Across currently deployed industrial AI-engineering projects, the typical improvement range for industrial enzyme catalytic activity is consistently between 2- and 10-fold, establishing AI as a core driver for the iterative upgrading of next-generation industrial enzyme products.

V. Emerging Industry Trends: Enzyme Catalytic Activity Officially Enters the Era of AI-Programmable Engineering

Looking across the complete evolutionary trajectory of the enzyme engineering industry, enzyme catalytic activity optimization has undergone a historic transition—from early natural strain screening and random artificial mutagenesis to AI-driven full-process precision programmable custom design, fundamentally reshaping the traditional R&D logic of enzyme engineering. At present, AI has established a complete closed-loop industrial enzyme optimization system covering "kinetic prediction → de novo molecular design → virtual screening":

At the prediction front: Deep learning models such as IECata can accurately predict core kinetic parameters (kcat/Km), enabling early elimination of inefficient sequences and substantially reducing wet-lab workload.

At the design front: Diffusion models such as RFdiffusion2 enable atomic-level precision customization of active sites, directionally amplifying core catalytic performance.

At the screening front: Machine learning enables direct prediction of enzyme activity, substrate specificity, and environmental tolerance from amino acid sequences, compressing the costs of large-scale screening experiments.

AI-enabled enzyme catalytic activity optimization has progressed through three technical advancement phases:

Phase 1: Assistive tool layer — limited to protein structure resolution and mechanistic prediction.

Phase 2: Generative engine layer — capable of zero-shot de novo construction of entirely new enzyme sequences with high catalytic potential.

Phase 3: Multi-objective synergistic optimization layer—simultaneously balancing multiple industrial core parameters, including catalytic activity, thermostability, substrate selectivity, and environmental tolerance, directly addressing the pain points of industrial translation.

Looking to the future, leveraging AI protein generation foundational technologies, enzyme catalytic activity will progressively evolve into a standardized, on-demand tunable programmable engineering property. Across diverse application scenarios—whether in biochemical fermentation, innovative drug synthesis, or clinical in vitro diagnostics—high-performance enzyme formulations tailored to specific operating conditions can be custom-designed through AI, effectively expanding the performance boundaries of natural enzymes, overcoming traditional engineering cost and capacity bottlenecks, and comprehensively advancing the large-scale, high-efficiency development of the green biomanufacturing industry.