AI Artificial Proteins: Unlocking Programmable Life Molecules and Transforming the Biomanufacturing Industry

Have you ever wondered that in the future, antibodies to treat cancer, enzymes to degrade plastics, and antigens for vaccines might not be "found" in nature, but rather "designed" by AI in computers?

This is not science fiction. The 2024 Nobel Prize in Chemistry was awarded to David Baker (computational protein design) and Demis Hassabis and John Jumper (protein structure prediction), marking the arrival of an era of AI-powered protein science. From proof-of-concept in the lab to large-scale mass production, AI-designed proteins are reshaping the underlying logic of the life sciences industry at an unprecedented pace, pushing protein R&D from the "agricultural era" to the "industrial age."

I. AI Artificial Proteins: Definition and Core Technical Logic

To understand AI artificial proteins, we first need to start with proteins.

Proteins are the main executors of life activities—from enzymes that digest food, to antibodies that fight viruses, to cytokines that transmit signals, almost all life functions are performed by proteins. Natural proteins are the product of billions of years of evolution in nature. In the past, humans could only "discover" and "modify" them, but it was difficult to "design from scratch."

AI artificial protein (AI-designed protein / AI-generated protein) refers to functional protein molecules that rely on artificial intelligence algorithms and are targeted or newly constructed through three technical paths: sequence optimization, structural remodeling, and de novo protein design. They belong to the category of artificially programmable life molecules and are fundamentally different from naturally synthesized proteins in living organisms. Among them, the route with the highest technical barriers and greatest development potential is de novo protein design, with the two R&D models having completely opposite logic:

Traditional protein engineering: Relying on existing natural protein → multiple mutation modifications → screening individuals with optimized performance

AI de novo protein design: Anchoring target functional requirements→ AI directly generates new amino acid sequences that match performance

For example: traditional protein engineering is like "renovating an existing house," while AI artificial protein is "drawing blueprints and building a new house according to your needs."

II. Natural Proteins Have Inherent Performance Limits, AI Design Is the Key to Breaking Through

Natural proteins have a "performance ceiling" that’s just too low, while human demands are way too complex. Here are a few real pain points:

Natural enzymes can’t handle high temperatures. In industrial production, many reactions need to happen at high temperatures, but natural enzymes start to lose activity above 60℃. Traditional methods require dozens of rounds of directed evolution, taking years just to improve thermal stability by a few degrees.

Natural Protein A can't withstand alkalinity. In antibody drug production, Protein A chromatography resins need to be cleaned on-site with 0.1–0.5 M NaOH after each use. Wild-type Staphylococcus Protein A easily deamidates and denatures under strong alkali; the industry spent over twenty years using genetic engineering mutations (like GE’s MabSelect SuRe) to develop a usable alkali-resistant version, but it still falls short for higher-concentration alkali solutions or more cycles.

Many targets don’t have natural ligands. In drug development, some "hard-to-drug target" proteins have surfaces with no obvious binding pockets, making it hard for traditional natural proteins to achieve targeted binding, leaving drug discovery at a dead end.

The root cause of all these problems is the same: proteins in nature evolved for "survival," not for "industrial production" or "human medicine."

To break through the performance limits of natural proteins, you have to design them from scratch. But why wasn’t this possible before?

Because the protein sequence space is enormous. A protein of medium length has 300 amino acids, giving 20^300 possible sequences—far more than atoms in the observable universe. Relying on human experience and random screening is like finding a needle in a haystack.

AI changes all of this. It doesn’t just "try," it "infers"—by learning from billions of natural protein sequences, AI grasps the "grammar" between amino acid sequences, 3D structures, and functions, allowing it to navigate the enormous sequence space precisely and directly find candidate molecules that meet specific performance requirements.

III. Three Core Practical Applications Covering Medicine, Industry, and New Materials

Three Major Commercial Applications 

Proteins designed by AI are moving from the lab to industry, covering everything from medicine to industrial applications.

1. Biomedicine: Faster, more precise antibodies and drugs

This is the most talked-about application of AI-designed proteins. Traditional antibody development requires animal immunization and library screening, which takes a long time and has a low success rate. AI can design binding proteins—antibodies, nanobodies, binders—directly targeting the structure of the target from scratch, producing high-quality candidate molecules in just a few weeks. According to industry data, several AI-designed antibody molecules have already entered clinical trials, with the fastest going from target selection to clinical advancement in less than 2 years, whereas the traditional path usually takes over 5 years. For disease treatment, the speed boost here is huge.

Besides antibodies, AI-designed therapeutic proteins like cytokines, growth factors, and fusion proteins are also progressing rapidly, aiming for higher activity, better stability, and lower immunogenicity.

2. Industrial enzymes: Green manufacturing "super catalysts"

Industrial enzymes are another major area for AI-designed proteins. They are widely used in food processing, textiles, detergents, bioenergy, plastic degradation, and more. But natural enzymes often only work under mild conditions, while industrial production demands "extreme conditions" like high heat, strong alkali, and high substrate concentrations.

AI can systematically improve enzymes—boosting heat stability, expanding pH tolerance, enhancing substrate specificity, and increasing catalytic efficiency. An AI-optimized industrial amylase that remains highly active at 60°C could cut a production step's steam cost by a third; an enzyme that can degrade PET plastic at room temperature could completely change the plastic recycling industry.

3. Functional protein materials and tools: Opening up new possibilities

AI can also design functional proteins that don’t exist in nature—like new structural proteins, fluorescent proteins, sensor proteins, protein nanomaterials, and more. These molecules might not have natural equivalents, but with AI, humans can "create from scratch" proteins with specific structures and functions.

The 2024 Nobel Prize in Chemistry was awarded to scientists in the field of protein design and structure prediction, highlighting breakthroughs in this direction. As a 2025 review in Nature Reviews Bioengineering pointed out, AI is transforming protein design from a trial-and-error process into a predictable and reproducible engineering discipline.

IV. Global Market Expansion and Domestic Intelligent Manufacturing Infrastructure Completion

Global capital is rapidly flowing into this sector. According to Mordor Intelligence, the global AI protein engineering market is expected to grow from $1.5 billion in 2025 to $4.75 billion in 2031, with a compound annual growth rate of 21.2%. Another agency predicts that the In Silico protein design market will increase from $1.87 billion in 2025 to $4.86 billion in 2030.

The underlying support for this rapid industry growth is the transition of AI protein technology from lab papers to large-scale production. Domestic hardware infrastructure has already formed a first-mover advantage: the National Protein Science Research (Shanghai) facility in Zhangjiang, Shanghai, has built the world’s first AI-driven, cell-free automated protein manufacturing platform, capable of fully synthesizing and testing tens of thousands of proteins per day. The loop from AI-designed sequences to real protein output can be completed within 24 hours, providing a hardware foundation for local companies to commercialize their products.

V. Opportunities in the Chinese Industry: Practice of the MatwingsVenus™ (Xiaowu™) Platform

 AI Closed-Loop Research and Development Process for Dry and Wet Applications

Chinese companies have not been absent from the global AI protein race. Represented by local forces like MatwingsVenus™ (Xiao Wu™) under Shanghai Matwings Technology, they are carving out a full-chain path of 'AI design → automated experiments → industrial application.'

The core idea of the MatwingsVenus™ (Xiao Wu™) platform is to turn protein research capabilities that were previously only accessible to large research institutions and top companies into an engineering platform that more companies and researchers can efficiently use. The platform is agent-centered, supports retrieval of billions of real-labeled protein data, integrates 200 professional protein design tools, 50 platform-certified experts, and 30 skills fine-tuned by experts in various fields. Users can put forward functional requirements in natural language, and agents on the platform will automatically break down tasks, call the relevant tools, and complete the full computational process from sequence design to performance prediction.

More importantly, the MatwingsVenus™ (Xiao Wu™) platform seamlessly links cloud-based AI design capabilities with automated wet-lab platforms. Designed sequences can directly drive robots to complete sample preparation, purification, and functional testing, achieving a one-stop closed loop of 'design—verify—iterate.'

So far, the platform has validated the practical effects of AI design in multiple areas:

· Industrial protein optimization: Matwings Technology collaborated with GeneScience Pharma to develop an alkali-resistant single-domain antibody in just four months. Its alkali resistance improved fourfold in extreme pH 13–14 conditions, lifespan doubled, and they successfully achieved 5000-liter scale industrial production. This product is the first protein product in the world designed by a large model that reached industrial production, saving the company over tens of millions of yuan annually, and was included in the Ministry of Industry and Information Technology's first batch of 'Typical AI Applications in Biomanufacturing,' receiving an 'Excellent' rating in 2025.

· De novo protein design: The platform can design binders for multiple types of biological targets, generating candidate proteins, peptides, nanobodies, cyclic peptides, etc., based on targets like proteins, peptides, small molecules, and more. In de novo binder design projects targeting immunoregulation, multiple molecules showed clear functional activity in in vitro experiments.

Matwings Technology’s rapid development has consistently earned recognition from the capital market. In March 2025, the company completed over 200 million yuan in Series A financing; in July 2026, it completed several hundred million yuan in another Series A round, co-led by Shanghai Guotou Pioneer and Guotou Innovation, with additional participation from multiple institutions and continued investment from existing shareholders. The financing will mainly be used for the continued R&D and iterative upgrades of the core MatwingsVenus™ (Xiao Wu™) platform, accelerating the large-scale application of AI-driven protein design in the circular economy, innovative drugs, healthcare, medical consumables, and other fields.

In November 2025, Matwings Technology also reached a strategic partnership with Bayer Health Consumer Products at the Import Expo, with both parties set to further deepen cooperation in AI technology for scientific research, new product development, and application transformation.

VI. Conclusion: Entering the Programmable Era of Protein Research

Human use of proteins has gone through three major leaps:

The first was discovery—finding useful proteins in nature, like insulin and trypsin;

The second was modification—optimizing natural proteins through genetic engineering and directed evolution;

The third is design—AI creating entirely new proteins from scratch based on functional needs.

We are standing at the starting point of this third leap. When proteins shift from being 'gifts of nature' to 'molecules that humans can design,' industries relying on biomolecules—biomedicine, industrial manufacturing, materials science, environmental protection…—could all be redefined.

For Chinese companies, this is both a challenge and an opportunity. As the research paradigm moves from 'experience-driven' to 'data-driven,' the barriers to early advantage are being redefined—whoever can more quickly build a closed loop of AI design and experimental validation will gain the upper hand in the new round of industrial competition.

The story of AI-designed proteins has just begun, but its endpoint is far beyond what we can imagine today.