Protein Toolbox: Installing a 'Thinking Operating System' for Protein Design

A super toolbox that combines AI prediction, automated experiments, and expert knowledge is turning protein design—once something only a few scientists could accomplish—into infrastructure accessible to more people.

Protein design, a high-end technology once reserved for a handful of structural biologists, is undergoing a profound transformation.

Why? Because traditional protein development methods are simply too 'heavy.' A researcher has to manually select scaffolds, adjust interface residues by hand, optimize sequences one by one, and wait weeks or even longer for wet lab feedback after each round of design. From concept to obtaining a usable molecule, it often requires multiple teams working in sequence, taking months or even years.

But as powerful protein design tools continue to emerge—from structure prediction to sequence generation, from functional annotation to pathway analysis—a paradox also arises: the more tools there are, the more fragmented the workflow becomes. To study a protein, you might need to check sequences and functional annotations in UniProt, look up structural information in PDB or AlphaFoldDB, analyze pathway relationships in Reactome, and then search protein interaction networks in STRING. Jumping from page to page, the information becomes increasingly scattered, and you still have to manually organize and repeatedly compare it—this is hardly a 'toolbox,' but rather a collection of disconnected parts each operating on its own.

What the market and industry are waiting for is an answer that can assemble these parts into an 'operating system.'

01

From 'tool collection' to 'intelligent brain'

AI BRAIN

A good toolbox shouldn’t make you figure out by yourself which to open first and which to open later. A true 'protein toolbox' should be like an experienced mentor—you only need to clarify your goal, and it takes care of planning the path, calling the tools, and integrating the results.

This is precisely the core evolution direction for protein R&D tools in the AI era.

This direction has already produced remarkable practical results. The MatwingsVenus™ (Xiaowu™) platform independently developed by Matwings Technology is a one-stop protein R&D platform centered around an 'intelligent agent.' Unlike traditional platforms that merely stack functions, its uniqueness lies in this: users only need to give task instructions in natural language, and the system will automatically break down the tasks, orchestrate the corresponding design, prediction, analysis, and screening capabilities, completing deep research, enzyme mining, directed evolution, and de novo design work.

You don’t need to remember how to use over 200 protein design tools individually, nor do you need to understand how more than 30 databases and over 400 professional tools coordinate behind a single instruction. You simply communicate with it like you would with a senior R&D colleague, telling it what you need, and it will compress weeks of manual database inquiries, comparisons, and design work into automated processes that take hours.

It’s like providing a protein designer with an 'AI co-pilot'—it doesn’t replace your judgment but carries all the repetitive and tedious tasks for you, allowing you to truly focus your energy on innovation.

02

A 'hands-on' toolbox: connecting the digital and physical worlds

Ten-billion-scale Protein Sequence Dataset

If we say that an intelligent agent's brain solves the question of "how to design," then the next challenge is even tougher: how to verify? In the past, the biggest pain point in protein design was the "separation of dry and wet processes." AI runs algorithm optimizations in the cloud, but the generated sequences still have to wait for lab scheduling—even if the AI runs extremely fast, the bottleneck of physical experiments always remains. As a result, the cycle of "design-wait-verify-redesign" was often dragged into a marathon.

A toolbox that can truly change the rules of the game must connect "thinking" and "doing." MatwingsVenus™ (Xiaowu™) is one of the platforms that takes a key step in this direction—it builds a "dry-wet closed loop": after AI completes a design, users can directly select "place an order" on the platform, seamlessly connecting to an automated shared laboratory to drive robots to carry out sample preparation, protein purification, and functional testing; experimental results are automatically fed back as a basis for the next round of AI design optimization, forming an iterative closed loop of "calculation-driven wet experiments and wet experiments feeding back to calculation."

This mode of "design as verification, verification as iteration" means that the previous disconnected process of "AI produces results, tosses them to the wet experiment team for verification, and waits weeks to see the effect" becomes an automated relay on the same production line. AI predictions for protein modification become more accurate with each round of wet experiments, and every experiment feedback makes the next design closer to the goal—like installing a "data flywheel" for protein design, which speeds up the more it runs.

03

Value through cases: when the toolbox actually works

Theoretical appeal ultimately cannot match the persuasiveness of practical results. Let us look at a real case.

Immune regulatory receptors are important targets in innovative biologic drug development and are widely used in areas such as oncology, autoimmune diseases, and inflammatory diseases. De novo design of this target is extremely difficult: due to novel target selection, there is a lack of similar drug molecules for reference; the target surface is dominated by polar regions and lacks typical high-druggability binding hotspots; and the natural ligand already has nM-level high affinity, further raising the threshold for candidate molecules to achieve effective blockade.

dozens of binder molecules

In such a "new and difficult" project, based on the MatwingsVenus™ platform ™, Matwings Technology takes the target structure and functional requirements as inputs, and the agent automatically completes the whole process of calculation such as skeleton screening, interface design, sequence optimization, and druggability prediction, and quickly outputs high-quality binder design sequences. Subsequently, the binder samples prepared by the automated experimental platform performed well in vitro cell viability experiments - dozens of new binder molecules with in vitro cell blocking activity were successfully obtained, with both functional inhibition and high affinity potential, and the whole process of designing binder from scratch was completed.

This is not a demo in the laboratory, but a practical implementation of AI-driven protein design capabilities on the most challenging innovative drug projects. And this case is just the beginning - there will be more and more such "toolbox-driven R&D" in the future.

04

From "big platform driven" to "personal availability": inclusiveness is the ultimate proposition

A truly great tool that will eventually lead to inclusiveness.

In the past, complete protein research and development capabilities - AI design, wet experiment verification, and expert consultation - were only played by big pharmaceutical companies and large institutes. Today, when a "super toolbox" that integrates more than 200 protein design tools, a database of tens of billions of tags, an automated shared lab, and more than 50 platform certification experts can be easily called by individual developers through natural language, the threshold of technology is being quickly smoothed.

This is not only an improvement in efficiency, but also a fundamental change in the R&D model - protein research and development from "big platform driven" to "personal availability". A PI in an academic lab, a founder of a synthetic biology startup, or even a graduate student working on functional protein materials may use conversational agents to complete closed-loop research and development that used to require the cooperation of a whole team.

In fact, Matwings Technology has clearly expressed this vision: in the future, it will continue to promote the inclusiveness of protein research and development capabilities, so that everyone can develop the protein they want. Such a vision is also supported at the capital level - following the completion of more than 100 million yuan of Series A financing in November 2024, Matwings Technology once again completed more than 200 million yuan of Series A+ financing in March this year, jointly led by PetroChina Kunlun Capital and Shanghai Future Industry Fund. The influx of real money confirms the market's confidence in the direction of "inclusive protein research and development".

05

Conclusion

Looking back at the history of protein engineering, in 1978, Canadian biochemist Michael Smith first proposed site-directed mutagenesis, marking the beginning of the era of human-modified proteins. Nearly half a century has passed, and protein design has evolved from site-directed mutagenesis to directed evolution, from physics-based rational design to AI-driven generative design, with each step lowering the barrier to "creating a new protein."

Today, we are witnessing the most crucial leap in this field: from "tools" to "toolkits," from "components" to "operating systems," from "the craftsmanship of a few" to "accessible infrastructure."

Whether you are engaged in innovative drug research, industrial enzyme design, the exploration of synthetic biology, or the development of functional protein materials, you can pay attention to the development trends of these new-generation protein toolkits—they may not think for you, but they will become your most powerful "AI R&D partners," helping you turn ideas into molecules and molecules into products.

Because the future of protein research and development will no longer be about who has the sharpest tools, but about who can more elegantly master a complete system of "thinking" tools.