How did proteins evolve from gifts of nature to AI-designed 'molecular assets'?

At the intersection of life sciences and business, 'proteins' are transforming from a purely biological term into a commercial buzzword driving trillion-dollar markets. From enzymes in laundry detergent to antibodies in innovative drugs; from the global trend of alternative proteins to a green revolution in industrial catalysis, proteins have long crossed the laboratory window to become the true center of competition in the industry chain. How can we understand the commercial value of proteins? And how is AI rewriting the industry logic around them? From the random gifts of nature to programmable molecules in the hands of AI, the value logic of proteins is being completely rewritten. This article will explore this with you.

1. Why are proteins valuable?

  Protein Value Chain

To understand the commercial value of protein, one must first recognize one fact: protein is nature's most sophisticated nanomachine.

In a commercial context, its value is reflected in three dimensions: in the pharmaceutical field, protein drugs (such as antibodies, cytokines, hormones) have become one of the fastest-growing sectors in the global pharmaceutical industry; In the industrial sector, enzyme preparations are widely used in food processing, laundry, textiles, bioenergy, and other industries. According to market research institutions, the global industrial enzyme market is expected to reach approximately 8 to 9 billion USD by 2026; In the food sector, alternative proteins such as microbial proteins are moving from the laboratory to the dining table, reconstructing the traditional protein supply system.

A protein molecule is composed of dozens to thousands of amino acids folded in precise sequences into a specific three-dimensional structure. The shape of this three-dimensional structure—the position of the active site, the distribution of binding domains, and the flexibility of conformational changes—determines its function: whether it cuts or connects, transports or conducts signals, whether it is recognition or catalysis.

This precise correspondence of "sequence → structure → function" gives proteins three unparalleled commercial qualities:

First, extreme specificity. An antibody protein can precisely recognize a specific antigen on the surface of cancer cells while "turning a blind eye" to normal cells—a selectivity difficult for small molecule drugs to achieve. According to recent global pharmaceutical sales rankings, antibodies and antibody-derived drugs occupy half of the top 10 spots for years, and 'drug kings' with annual sales exceeding $10 billion are constantly emerging."

Second, ultra-high catalytic efficiency. Industrial enzymes can increase chemical reaction rates by hundreds of thousands or even millions of times, and operate at normal temperature and pressure without the need for high temperature, high pressure, or organic solvents. After AI engineering modifications to improve stability and substrate spectroscopy, industrial enzymes can operate in wider temperature ranges and pH windows, replacing chemical process routes that previously relied on high temperature, high pressure, and heavy metal catalysis, significantly reducing energy consumption and hazardous waste generation.

Third, programmability. In principle, changing the amino acid sequence can alter the three-dimensional structure and function of proteins. In other words, proteins can be "designed." This "programmability" was once extremely underestimated, but AI is fully activating it.

2. From "Discovering Proteins" to "Designing Proteins"

AI is revolutionising traditional Research and Development

If we were to draw a dividing line in the history of protein technology, it would be: from discovery to design. 

Our ancestors were using proteins thousands of years ago—brewing, making cheese, tanning leather—using the enzymes nature provided. By the 20th century, with protein sequencing, X-ray crystallography, and later DNA sequencing, scientists began to "see" protein structures and understand how they work. Then came directed evolution (which simulates natural selection in the lab by using random mutations and screening to accelerate protein optimization), allowing researchers to mimic natural selection and quickly screen for improved enzyme variants in the lab—the 2018 Nobel Prize in Chemistry was awarded for precisely this method. 

But all these approaches are essentially "discovery" logic: starting from proteins found in nature and making random tweaks or just screening from there. True "design"—calculating from scratch which amino acid sequence is needed for a specific function—was impossible in the past. 

Take a medium-length protein of 300 residues, for example. With 20 natural amino acids, the possible combinations for a 300-residue protein are 20 to the power of 300—which is far more than the number of observable atoms in the universe. Trying to "design" by human intuition or brute force would be like finding a needle in a haystack. 

Traditional protein R&D has long faced two main bottlenecks: first, the vast sequence space makes the traditional trial-and-error method extremely inefficient; second, the gap between computational design and wet lab experiments creates a long "manual relay," making the process disjointed and error-prone. 

This is exactly where AI changes the game. 

A 2024 review published in *Biomolecules* points out that protein language models, structure-based design, and machine learning integration have significantly improved the ability to predict protein properties and guide engineering modifications. Deep neural networks, trained on massive amounts of protein sequence and structural data, have learned the "grammar" and "vocabulary" between amino acid sequences and 3D structures, efficiently navigating the huge sequence space to find candidate sequences that meet specific performance constraints. 

In other words, AI isn’t "trying"—it’s "reasoning." Based on learned protein "design principles," it predicts how a sequence will fold, what functions it might have, and whether it will be stable under high temperatures or certain pH levels. 

3. The Three Major Commercial Battlegrounds for AI Protein Design

 Commercial Applications

After understanding the technical principles, a more important question is: In which areas is AI protein design actually generating real commercial value? The answer can be narrowed down to three core battlefields.

Battlefield One: Industrial Enzymes – The Invisible Infrastructure Supporting Trillion-Dollar Manufacturing

Enzymes are the core catalysts of biomanufacturing, covering almost all industrial fields from food processing, textiles, and paper making to detergents and biofuels. The traditional R&D model for enzymes is 'find an enzyme in nature → tweak a bit in the lab → scale up for production,' which usually takes 3 to 5 years, and the resulting enzymes are often just 'good enough,' far from optimal.

AI is shortening this cycle by an order of magnitude.

Take the most common industrial demand—improving enzyme thermal stability—as an example: A 2025 review in the World Journal of Microbiology & Biotechnology analyzed the molecular determinants of enzyme stability in detail and pointed out that AI-assisted enzyme design is turning the development of 'super enzymes' that are heat-, acid- and solvent-resistant from 'happy accidents' into 'predictable design.' A 2026 review in Molecules further explained that structure prediction models like AlphaFold2 and RoseTTAFold can provide high-precision 3D coordinates, which, combined with protein language models like ProGen and ESM-2 that infer stability and functional sites, along with downstream tools like active site prediction and molecular docking, can greatly improve the predictive accuracy of enzyme catalytic functions. Generative models can even create synthetic enzymes from scratch with custom catalytic activity—so-called synzymes.

What does this mean in practice? Take textile desizing, for example. Traditional processes require treating fabrics in near-boiling alkaline solutions, which consumes a lot of energy. A starch-degrading enzyme optimized by AI to remain highly active below 60°C can directly cut steam costs for this step by a third. Across the massive global industrial manufacturing system, such 'single-point optimizations,' when accumulated, release enormous economic value and carbon reduction effects.

Battlefield Two: Therapeutic Proteins — From Antibodies to Peptides, the Logic of New Drugs

Protein drugs—especially monoclonal antibodies—are the absolute backbone of today’s pharmaceutical industry. As of August 2025, industry research shows that 144 antibody drugs have been approved by the FDA worldwide, and over 1,500 candidate molecules are in clinical development. According to sources like Evaluate Pharma, the global antibody drug market in 2024 is estimated to be around $200–220 billion. But traditional antibody discovery relies on animal immunization or large-scale library screening, which is time-consuming and has a low hit rate. The lead molecules obtained often require extensive subsequent engineering modifications.

A 2024 review published in *Biomolecules* pointed out that the integration of computational and experimental methods is driving the development of next-generation biologics, including affinity maturation, bispecific antibodies, enhanced enzyme stability, and conditionally active cytokines. AI tools like protein language models can "learn" from natural antibody libraries to see which CDR (complementarity-determining region) sequences are most likely to have high affinity for a specific target. Then, millions of virtual variants can be generated and evaluated computationally, with wet lab validation done only on the most promising few—this is the "dry-wet closed loop".

The trend is equally clear in the peptide drug field. The global peptide therapeutics market surpassed $50 billion in 2024. Industry stats show that about 100 peptide products are approved in major markets, nearly half of which emerged in the last 20 years. AI-driven peptide design can maintain target affinity while predicting and optimizing stability, oral bioavailability, and immunogenicity—all traditional bottlenecks in peptide drug development.

Battlefield Three: Sustainable Materials — Proteins as "Green Building Blocks"

Beyond catalytic and therapeutic functions, proteins also have an emerging commercial application: as materials themselves.

Structural proteins like spider silk, silkworm silk, and elastin-like polypeptides (ELPs) can have their mechanical properties, degradation rates, and biocompatibility tuned via AI design, making them alternatives to petroleum-based plastics and non-degradable synthetic fibers. In high-end applications like sneaker midsoles, medical sutures, and absorbable implants, AI-designed protein materials are moving from lab samples to commercial-scale production.

4. MatwingsVenus™ (Xiaowu™): Giving Protein R&D an “Intelligent Agent”

The common prerequisite for the three major commercialization battlegrounds mentioned above is the need for an AI platform that can efficiently complete "sequence → structure → function" prediction and optimization.

On April 24, 2026, Shanghai Matwings Technology officially launched the conversational protein research and development AI, MatwingsVenus™ (Xiaowu™). Unlike traditional tool platforms, it builds a one-stop protein R&D system centered around the AI agent — users input task goals in natural language, and the system automatically breaks down tasks, schedules tools, and organizes workflows, instead of merely piling up functional modules.

In terms of core capabilities, the platform supports retrieval of tens of billions of real-labeled protein data, integrates 200 protein design tools, 50 platform-certified experts, and 30 skills tuned by experts in various fields, covering deep research, enzyme mining, directed evolution, de novo design, and automated wet lab collaboration.

Regarding query capabilities, the platform has access to 30 database modules and 400 specific tool functions, covering multiple dimensions such as protein sequence, structure, functional annotation, pathway analysis, interactions, and expression analysis, supporting parallel searches across multiple databases with automatic result integration.

For design models, the platform update in June 2026 introduced three core models to MatwingsVenus™ (Xiaowu™): BoltzGen, LigandMPNN, and Protenix. BoltzGen enhances de novo design of new binding proteins and can generate candidate binding molecules for proteins, peptides, and small molecule targets; LigandMPNN focuses on precise sequence design for systems with small molecule ligands or metal ions; Protenix predicts the 3D structure of multi-component systems including proteins, DNA/RNA, small molecule ligands, and ions, and analyzes intermolecular binding conformations and interaction interfaces.

The most crucial feature is the "dry-wet closed-loop" model — the platform connects the full chain from AI design to automated experimental verification. After the agent completes the design, it can directly link to automated shared labs for execution, driving robots to handle sample preparation, protein purification, and functional testing, with experimental results fed back into the next round of AI design, forming a conversational dry-wet iterative loop where "computational design drives wet experiments, and wet experiments inform computation."

In practical application, according to Matwings Technology’s public disclosure, the company has delivered 40 protein design projects for leading clients across sectors including innovative drugs, in vitro diagnostics, industrial enzymes, and nutrition & health. In a de novo design project targeting an immune regulatory receptor, the platform successfully generated dozens of new binding molecules with in vitro cellular blocking activity, completing the full validation process for de novo designed binders. In a project with a listed pharmaceutical company, the team revamped key raw materials in less than a year, improving alkali resistance by four times, and scaling production to 5,000 liters per batch, saving the company tens of millions in protein raw material costs annually.

In May 2026, MatwingsVenus™ (Xiaowu™) Enterprise Edition was officially launched. Building on the one-click ordering for AI protein design and wet lab experiments available in the personal version, it added an intelligent management backend. In June 2026, Matwings Technology was included in 36Kr's "Top 100 Most Valuable Growth Companies of 2026," ranking in the artificial intelligence/large model track.

More importantly, this capability is being modularized and standardized: protein design is no longer a workshop-style task requiring top computational biology experts to engage deeply in each "individual project." Instead, it has become a platform service that R&D teams across different industries and scales can "call on."

5. The Critical Point of the Protein Economy

Looking back at history, human use of proteins has gone through four stages:

Protein 1.0 — Directly using natural proteins (brewing, fermentation); Protein 2.0 — Discovering and extracting specific proteins (enzyme industry, antibody drugs); Protein 3.0 — Directed evolution and semi-rational design, making tweaks on natural proteins; Protein 4.0 — AI-driven de novo design.

We are now on the threshold of Protein 4.0. In this stage, proteins are no longer just "natural resources" to be discovered, but "molecular assets" that can be designed, produced, and optimized on demand.

A forward-looking article published in *Cell Systems* in 2025 described this future vision: "The scientific community has so far uncovered only a tiny fraction of the enzymes generated by evolution, and AI methods are opening a door to an enzyme universe that goes beyond biological evolution — possibly even providing a path to genetically encode almost any chemical reaction."

When protein design becomes predictable, scalable, and reproducible, it will permeate almost all molecule-based economic activities — including pharmaceuticals, chemicals, materials, energy, food, and environmental protection — as a foundational technology platform. Proteins will no longer be a biological term alone — they are becoming an engineering discipline.

And one of the core laws of engineering is this: what can be designed, will eventually be mass-produced. Once AI becomes the "designer" of proteins, the imaginative space of the bioeconomy will only just begin to open.