MoleculeMind Breaks New Ground in AI-Powered De Novo Enzyme Design, Unveils SENZ at ICML 2025

Recently, artificial intelligenceTop Academic Conference in the FieldICML 2025Announcement of the Review Results for This Year's Papers.

By "AI ProteinFoldFounder”Professor XU JinboFounded AI Protein Design LeaderMoleculeMind(MoleculeMind) Joint Research Achievements with Hong Kong Polytechnic University"Retrieval-Augmented Zero-Shot Enzyme Generation for Specific Substrates"Included in the conference proceedings.

This study has achieved pioneering success inNatureNever seen beforeMolecular or biological reactions，Generate Custom Catalysts on Demand, and outperforms natural enzymes and enzymes designed by traditional methods in key indicators such as catalytic efficiency and stability, offering significant value for the development of industries like biomedicine and biomanufacturing.

Enzyme Production on Demand: Breaking the Trillion-Dollar Bioeconomy Development Bottleneck

Enzymes as highly efficient and environmentally friendly in nature"Molecular machines" are the core engine driving the development of the trillion-dollar bioeconomy in modern biomedicine, green chemical industry, environmental degradation, and green agriculture.

However, natural enzymes that have evolved over hundreds of millions of years are only suitable for reactions known in nature. The ever-emerging artificial chemical molecules, such as novel plastics, specific drug intermediates, and hard-to-degrade pollutants, have far exceeded their capacity limits.

The limitations of enzymes have become a bottleneck that urgently needs to be broken in the field of biomanufacturing. According to survey data from McKinsey,"Lack of Ideal Biocatalysts"It is the main obstacle to the scaled production of the biotechnology industry. The annual capacity loss caused by enzyme limitations in just the pharmaceuticals, chemicals, and agriculture sectors has already exceeded tens of billions of US dollars.

Traditional enzyme discovery and optimization methods, such as directed evolution or rational design, take months, are costly, heavily rely on expert experience, and have a low success rate.1%, and when faced with an entirely new substrate, it is almost "helpless," failing to meet the needs of the industry.

Emerging in recent yearsAI Protein Design Brings Dawn to Precise Enzyme Generation.AI methods can directly generate new catalysts by learning the structure-function relationships of a large number of known enzymes, making it possible to design enzymes with specific catalytic functions from scratch.

But because AI training highly relies on known enzyme-substrate pairing data, when faced with entirely new synthetic molecules, AI models often fall into a "blind spot" due to a lack of training data.

"But the real industry pain point is not just about optimizing known enzymes; it's about 'creating something from nothing' to design ideal new enzymes."Professor Xu Jinbo, founder of MoleculeMind, pointed out. MoleculeMind has conducted many years of research on AI enzyme optimization design and has made significant progress in novel enzyme design.

AI Zero-shot Enzyme Design: From Natural Evolution to Entirely New Creation

The core challenge in generating new enzymes, namely creating enzymes without direct catalytic data,MoleculeMindUnitedThe Hong Kong Polytechnic University, skillfully integrating bioinformatics big data retrieval with generativeAI, Innovatively Proposes a New AI Enzyme Design Method“SENZ”.The relevant paper has been accepted by the top AI conference ICML.

SENZ abandons the current mainstream protein similarity-based generation methods and innovatively retrieves and creates functionally relevant enzymes based on substrate structural similarity, establishing a design channel that goes straight from "unknown substrates" to "super enzymes."

The method guidesAI from"Similar Molecules"Deconstruction in the Catalytic CodeCore规律, Reorganized to Conquer New GoalsSubstrateThe"Molecular Key", PossessesThreeHeavy Innovation Mechanism: Based onA vast global enzyme database that identifies known enzymes which, although unable to directly catalyze the target molecule, have substrates structurally similar to the target molecule, providing a basis for enzyme design."Structural Blueprint"; then, under the precise guidance of an original enzyme-substrate classifier, summarize the underlying rules of biological reactions to construct a "Biological Reaction Atlas"; finally, design entirely new enzyme proteins that can efficiently and accurately catalyze target substrates with the help of generative AI.

The research team usedSENZ Designed a Non-Existent Pollutant Nemesis in Nature to Validate SENZ's Advancement.Methyl phosphonate is a refractory environmentalPollutants, no effective degradation methods have been found to date andEfficientNatural degradation enzymes. The team willThe design result of SENZ is based on Transfomer andProteinUnconditional Generation Method and Structural Generation Method of Language Modelsetc.A comparison of various mainstream enzyme design methods shows that the data indicates,Enzymes generated by SENZ significantly outperform baseline and natural enzymes.

"This proves that SENZ can not only mimic nature but also design solutions that nature has not evolved," the Hong Kong Polytechnic collaborative team stated. The success of SENZ is equivalent to providing each chemical molecule with its own exclusive key, making bio-manufacturing potentially free from reliance on natural evolution.

InPharmaceutical Field, can rapidly design efficient enzyme-catalyzed pathways for complex and difficult-to-synthesize drug molecules, significantly reducing production costs and accelerating new drug launches; inEnvironmental Protection Field, can be tailored to create "super enzymes" that efficiently degrade stubborn pollutants such as plastics, providing a biological solution for environmental governance. InBiomanufacturingFields, can empower the green and efficient production of bio-based materials, fine chemicals, food additives, etc.

SENZ is just MoleculeMindDesignThe"One of the 'biological keys'."Currently, MoleculeMind has developed more than ten innovative and industrially valuableNew Method for AI Protein Design, Including the World's First Multimodal AI Protein Foundation ModelNewOrigin`, and the industry's first fully-featured, one-stop AI platform for protein prediction, optimization, and design.`MoleculeOSetc.

Relying on these cutting-edge technologies, MoleculeMind hasCathay BiotechDeep collaboration with more than ten leading enterprises in the biomanufacturing and biopharmaceutical fields to develop urgently needed industry solutions."Super protein." For instance, in the field of biomanufacturing, MoleculeMind and Cathay Biotech jointly optimized a key enzyme protein. Compared to the wild strain, an AI-designed protein structure increased the strain's productivity by five times, which is expected to further enhance the commercial profitability of the product.

In the field of drug research and development, MoleculeMind collaborates with pharmaceutical companies to tackle the challenge of protein vaccine stability using AI.AI Designs Dozens of Ideal Candidate Proteins in Just Three Days, and exhibited higher neutralizing antibody titers and stronger cellular immunity, while breaking through the stability patent of the relevant vaccine.

"In the next three years, we will extend our 'design-on-demand' capabilities to the entire fields of antibodies, vaccines, and industrial enzymes," revealed Xu Jinbo. MoleculeMind will continue to forge a new AI engine for designing biomolecules, providing innovative 'biological solutions' to address major challenges in health, environment, and sustainable development.