Breaking Free from Fragmented R&D: AI Research Agents as Dedicated Executors in Protein Design

In today’s fast-paced development of synthetic biology and innovative drug research, protein R&D has long struggled with industry pain points such as fragmented tools, a split between wet and dry experiments, high manual labor costs in processes, and long development cycles. In the traditional model, researchers need to constantly switch between more than ten independent software, databases, and experimental systems, spending a lot of time on repetitive tasks like data export, format conversion, and cross-platform integration, which seriously slows down innovation in molecular modification and target protein development. With the deep integration of AI into scientific research, a completely new R&D paradigm centered around research intelligent agents is reshaping the underlying logic of the entire protein R&D process.

From a broader perspective, this transformation is not an isolated event. According to industry research agencies, the global AI for Science market is expected to reach about $4.538 billion by 2025, growing to $26.23 billion by 2032, with a compound annual growth rate (CAGR) of 28.9%. Within this, the global AI protein design market is expected to be about $496 million in 2025 and reach $1.494 billion by 2032. The market for agent-based AI in scientific discovery and research is projected to grow from $260 million in 2025 to $400 million in 2026, with a CAGR as high as 57%. These figures clearly point to a trend: AI-driven research tools are upgrading from “efficiency aids” to “R&D infrastructure,” with research intelligent agents at the core of this trend.

In April 2026, Matwings Technology launched the conversational protein R&D intelligent agent MatwingsVenus™ (Xiao Wu™). This platform uses an agent as the core, allowing users to input R&D goals in natural language, and the system automatically breaks down tasks and coordinates sequence design, structure prediction, performance analysis, and molecular screening. It offers a one-stop solution for in-depth literature mining, industrial enzyme discovery, directed protein evolution, de novo molecular design, and fully automated wet experiments working collaboratively. The platform connects cloud computing design with physical experimental workflows, completing a full closed-loop R&D process from literature and patent research, protein molecular design to automated experimental verification.

Computation-to-Experiment Workflow

1. Traditional Protein R&D Bottlenecks: Fragmented Processes Slow Down Industry Innovation Efficiency

At present, protein research in China generally relies on multiple sets of tools manually stitched together, with the full R&D process chained by human effort, resulting in numerous breakpoints throughout the chain. Sequence analysis, protein structure prediction, site-directed mutation design, and in vitro functional verification are handled by separate systems. Researchers have to manually transfer data, adapt formats, and coordinate outsourced experiments repeatedly.

This manual chaining approach leads to two major industry pain points: first, R&D heavily depends on the experience of senior staff, making it slow for newcomers to learn, and molecular screening trial-and-error costs remain high; second, dry computing and wet lab experiments are long separated, so high-quality protein sequences designed by AI cannot be quickly validated, leaving many promising candidate molecules unused. Traditional protein modification projects often take several years, and establishing a complete R&D system requires significant computing power and investment in automation equipment, making high-quality de novo protein design very challenging for small labs and startup research teams.

From an industry-wide perspective, this issue is becoming a structural bottleneck that limits efficiency improvements in biomanufacturing and innovative drug R&D. The state has already noticed this trend. In early 2026, the National Supercomputing Internet Platform officially launched a Scientific Computing Intelligent Agent, which can automatically handle research task breakdown, computing resource scheduling, software invocation, and result analysis via natural language interaction, reducing tasks that traditionally take a full day to just about an hour. This indicates that scientific intelligent agents, as a new generation of R&D infrastructure, are moving from exploratory lab use toward large-scale application.

2. Core Transformation of the Scientific Intelligent Agent: From Passive Tool to Full-Chain R&D Executor

Traditional AI protein tools can only respond to single-step commands, whereas the MatwingsVenus™ (Xiaowu™) scientific intelligent agent has natural language understanding, autonomous task decomposition, and holistic resource scheduling capabilities. Its core innovation is breaking the shackles of fragmented R&D. Researchers don’t need to master multiple professional software tools; they can simply express their R&D needs in natural language, and the intelligent agent will autonomously generate a complete workflow, performing literature review, molecular AI design, multi-round computational simulation, and automated experimental implementation all at once.

The platform builds a one-stop protein R&D foundation centered around the intelligent agent, integrating billions of protein datasets with real functional labels, over 200 professional protein design and analysis tools, more than 50 certified field experts on the platform, and over 30 specialized skills fine-tuned by domain specialists. This complete system connects the entire chain from research ideation to industrial process scale-up, shifting protein R&D from manual trial-and-error to intelligent, standardized, and engineering-oriented operations.

3. Three-Layer Progressive Capability Framework to Build a Self-Operating Intelligent R&D Loop

Tiered Workflow for Biotech AI Task Execution

MatwingsVenus™ (XiaoWu™) relies on a three-layer, stepwise linked capability system to achieve fully autonomous operation of the entire protein R&D process, which is also the core differentiator from single-function AI tools on the market.

3.1. Intelligent understanding of requirements and autonomous task breakdown

The platform supports natural language interactive R&D. Researchers input customized requests, like 'design heat-resistant industrial proteases' or 'optimize target-binding proteins to improve druggability.' The AI automatically parses the core R&D goals, breaking them down into multi-level sub-tasks such as literature and patent search, target 3D structure analysis, protein scaffold selection, sequence physicochemical optimization, performance prediction, and standardized experimental plan arrangement. It then automatically generates a complete R&D workflow, replacing traditional manual process planning.

3.2. Automatic allocation of resources and batch computation execution

The AI can automatically match the optimal data, algorithms, and expert resources based on the R&D scenario. It can independently complete the full set of computations, including generating new sequences, protein structure simulation, interface optimization, physicochemical property evaluation, and druggability prediction, without researchers needing to switch systems or manually transfer data.

3.3. Integration of wet and dry experiments with AI autonomous iterative optimization

This is the core technological breakthrough of the MatwingsVenus™ (XiaoWu™) AI. Traditional computational tools only handle cloud-based virtual design. MatwingsVenus™ (XiaoWu™) bridges the gap between virtual design and the physical lab: after the AI completes molecular design, it uses the platform's self-developed communication architecture to automatically sync sequence data with plasmid ordering and automated experimental scheduling systems, driving lab robots to perform protein sample preparation, affinity purification, and in vitro functionality testing.

4. Conversational wet-dry closed loop: greatly reduces protein R&D screening time and costs.

Unified Biotech Collaboration Hub

The disconnect between computational and experimental work is a core bottleneck limiting the protein R&D industry. The gap between virtual design and physical experiments directly leads to uncontrollable development cycles, low efficiency in screening positive molecules, and high overall R&D costs.

MatwingsVenus™ (Xiaowu™) has developed its own conversational closed-loop system that connects digital computation with physical experimentation. The full process is: user inputs functional requirements → the AI agent autonomously completes the full AI molecule design → automatically issues automated experimental tasks → robots perform activity testing in bulk → experimental data feedback drives AI iterative optimization.

In a project designing immune-regulating receptor targets from scratch, the platform uses the target’s 3D structure and functional requirements as inputs. The agent autonomously completes backbone screening, interface design, multiple rounds of sequence optimization, and preliminary druggability assessment based on computational models, ultimately screening dozens of novel binder molecules with clear in vitro cell-blocking activity. The entire R&D process can be completed through dialogue between researchers and the AI agent, without multiple teams coordinating across stages.

5. Clear human-AI collaboration division: AI handles repetitive tasks, researchers focus on original innovation

The role of the research AI agent is as an assistant, not a replacement for researchers. This new human-AI division maximizes the value of R&D resources.

All standardized and repetitive tasks—literature search and integration, multi-tool scheduling, batch computation, automated experimental collaboration, and experimental data review and iteration—can be efficiently handled by the AI agent. According to publicly available data from Matwings Technology, protein design timelines have shrunk from the traditional 2–5 years to 2–6 months, significantly improving R&D efficiency.

Meanwhile, proposing original scientific questions, evaluating research directions, and controlling academic and commercialization decisions—tasks that heavily rely on professional intuition and creative thinking—remain under the researchers’ guidance. The AI handles all the routine work, allowing the R&D team to focus on upstream innovation and core scientific breakthroughs.

From an industry evolution perspective, the role of research AI agents is shifting from "task executors" to "R&D collaborators." At this stage, AI research tools mainly focus on task execution. In the longer term, AI agents will gradually be able to independently propose scientific hypotheses and design validation paths, becoming true collaborators in the research process. As demonstrated by national supercomputing internet science AI agents, they can complete task decomposition and execution in tens of minutes that would take researchers hours—human-AI collaboration is moving from concept to everyday reality.

6. Inclusive R&D Infrastructure: Lowering the Entry Barrier for Advanced Protein Innovation

In the past, full-chain, high-precision protein R&D systems were only within the reach of top pharmaceutical companies and large research institutes. Small and medium teams or university labs were limited by computing power, specialized tools, and the heavy asset requirements of automated lab equipment, making it hard for them to engage in cutting-edge de novo protein design or directed evolution pipelines.

Matwings Technology leverages MatwingsVenus™ (Xiaowu™) research intelligence to democratize industry infrastructure, turning the complex R&D capabilities that used to be exclusive to big institutions into a "shared lab" that even individual developers can easily access. Various research entities can call upon top-tier AI models, full toolchains, and standardized automated lab resources anytime without investing huge sums in setting up labs and computing clusters, allowing them to independently carry out high-quality protein engineering and validate innovative molecules.

From scattered tools that require manually switching between multiple software programs to a conversational research intelligence that can autonomously plan and coordinate experiments, the protein R&D industry is officially moving past the trial-and-error phase driven by experience and entering a new era of smart R&D with dialogue interaction, closed-loop iteration, and industrialized production. In the future, MatwingsVenus™ (Xiaowu™) will continue to iterate on its core intelligence capabilities, empowering breakthroughs in therapeutic protein development, industrial enzyme biomanufacturing, synthetic biology, and more, helping the domestic protein research industry shorten R&D pipelines, reduce innovation costs, and accelerate the practical application of original protein technologies.