Bio R&D Agents: The R&D Revolution from Trial and Error to Prediction

In the life sciences field, developing a new drug often costs billions of dollars and takes more than a decade. Nearly 90% of candidate drugs fail during clinical development, and the average cost of bringing a new drug to market has reached tens of billions of dollars. This doesn't reflect a lack of scientific talent, but rather bottlenecks in traditional R&D methodologies—we're still pushing forward high-risk scientific exploration through a 'trial-and-error by hand' approach.

However, a profound shift in the model is happening. The rise of Agentic AI is transforming life science research from a 'R&D mode' to a 'research and prediction mode,' turning drug development from a high-risk gamble into a more predictable scientific process.

1. Agentic AI: From 'Executor' to 'Scientist'

AI-powered drug discovery

Compared to traditional generative AI, agent-based AI represents a fundamental evolution in artificial intelligence, shifting from passively executing instructions to actively identifying problems, planning, and carrying out solutions. It’s no longer just a tool for making predictions—it’s an autonomous system capable of reasoning, planning, and executing complex workflows on its own.

In the life sciences field, this means AI agents can play the role of a 'chief designer,' intelligently orchestrating the entire drug design process, optimizing R&D workflows, and assisting with experimental planning. There are research teams that have published work in the top academic journal Nature, showing how AI agents in virtual labs designed new nanobodies. The AI agents assembled a virtual scientist team including a 'chief researcher,' an 'immunologist,' and a 'machine learning expert,' successfully designing candidate nanobodies targeting COVID-19 variants.

2. 'Conversational Dry-Wet Loop': The Core Breakthrough of Biological R&D Agents

Dry and Wet Closed-Loop Video

Understanding bio-research intelligent agents comes down to grasping the core concept of the 'dry-wet loop.' In the industry, research often splits into dry computing (like digital modeling, sequence design, data analysis, virtual screening, etc.) and wet experiments (like sample preparation, molecular synthesis, activity testing, functional characterization, and other hands-on experiments). In traditional models, there's a big gap between computer-based design and experimental validation—after scientists finish the calculations, they have to manually transfer sequence information, reach out to outsourcing labs for synthesis and testing, making the whole process slow and error-prone.

The biggest breakthrough of bio-research intelligent agents is bridging these two worlds. Take Matwings Technology's conversational protein research agent, MatwingsVenus™ (Xiaowu™), for example: users just input task goals in natural language, and the system automatically breaks down the tasks, scheduling the necessary design, prediction, analysis, and screening capabilities. It can handle in-depth research, enzyme mining, directed evolution, de novo design, and even automated wet experiment coordination. The platform integrates over 200 protein design tools, billions of real labeled protein data points, and more than 50 platform-certified experts. Once AI finishes a design, the platform automatically feeds the results into plasmid ordering and experiment planning, driving robots to prepare samples, purify proteins, and check functions. The experimental results then flow back into the next round of AI design, creating a 'conversational dry-wet loop' where computation drives wet experiments and wet experiments refine computation.

This loop enables an intelligent R&D model of 'design as validation, validation as iteration,' deeply blending AI's predictive power with automated execution capability.

3. From Target Discovery to Molecular Design: How Agents Empower the Entire Chain of Biological R&D

The capabilities of bio-research agents cover the entire chain from early target discovery to final molecular optimization.

During the target discovery phase, multi-agent systems can collaborate to uncover large-scale original biological hypotheses. Currently, the industry's advanced multi-agent framework can integrate hundreds of professional tools and databases in the biomedical field, enabling automated information extraction and association analysis of thousands of documents and multimodal data in a single in-depth research model. Taking liver cancer target research as an example, under the supervision of human scientists, the entire chain from data mining and mechanism analysis to potential new target targeting was completed, and corresponding experimental validation protocols were automatically designed. This capability marks the evolution of AI agents from auxiliary tools to "virtual scientists" with comprehensive research execution capabilities.

Entering the molecular design stage, AI agents play a more central role. The agent can not only autonomously complete all computational tasks such as skeleton screening, interface design, and sequence optimization, but also significantly improve design efficiency through a closed-loop process of "AI-driven—experimental verification."

During the experimental verification stage, the agent system can autonomously arrange experimental tasks and drive the robot to perform high-precision operations. The automated experimental platform can reduce the traditional workload of "tens of thousands of tests" to "ten levels," achieving an extremely high positivity rate with very few tests. Existing studies have shown that by combining AI with robotics, multiple rounds of directional optimization of enzymes can be completed within four weeks, with each round requiring only the construction and characterization of hundreds of variants to achieve dozens of times performance improvement.

From practical implementation cases, the MatwingsVenus™ platform ™ has completed multiple validation projects in the fields of innovative drugs and synthetic biology. In the de novo design project for immune regulatory receptor targets, the platform successfully obtained dozens of novel molecules with in vitro cell activity, completing the full-process validation of the original design; In the sweetness protein transformation project, after multiple rounds of "agent design—automated experiments—AI feedback—redesign," the sweetness of several samples increased more than tenfold compared to the wild type.

4. Making R&D Capabilities More Accessible: The Industry Significance of Bio-R&D Intelligent Agents

The significance of bio-R&D intelligent agents lies not just in the technology itself, but in how they are reshaping industrial R&D capabilities. They are transforming biological research from complex skills once limited to large institutions into a kind of 'shared lab' that individual developers can easily access, enabling rapid conversion from ideas to products.

Take MatwingsVenus™ (XiaoWu™) as an example. Individual users can access AI design capabilities and automated wet lab platforms on the platform, completing the entire cycle from model simulation to experimental validation to iterative optimization through natural language dialogue. This means that what previously required multiple teams to complete high-threshold R&D processes can now be done by a single person on the platform. This technological revolution in research organization is bringing once high-cost, hard-to-access scientific skills down to earth, inspiring more personalized innovation from individuals and small teams.

5. Looking Ahead: A New Bio-R&D Model Driven by Intelligent Agents

AI agents are driving a revolution in the life sciences

Looking ahead, AI agents will further drive the life sciences field toward a new model of 'research and prediction.' By combining predictive results and digital simulations, we can significantly cut down the time and capital wasted on candidates that are bound to fail. Industry experts generally predict that by around 2026, we’ll see complete drug development pipelines driven by AI prediction at their core, with deep integration of predictive technologies across the entire chain—from early target screening and molecular design to preclinical validation—minimizing ineffective R&D investments.

AI agents will also reshape regulatory compliance processes, integrating compliance into workflows from day one and shifting human roles from clerical cross-checking to high-level strategic review. Data quality and AI readiness will be key factors in determining transformation success, so companies will need to build data pipelines that are accurate, consistent, complete, and well-structured.

Biology will eventually become a programmable engineering discipline. When AI agents can independently carry out the entire research process—from literature review to experimental validation—when computational design can directly drive automated experiments, and when every researcher has a 24/7 AI assistant, the jump in life sciences research efficiency is moving from vision to reality.

At the forefront of this change, platforms like the conversational protein R&D AI, MatwingsVenus™ (XiaoWu™), are turning the 'dry-wet closed loop' from a concept into actionable R&D infrastructure. From 'design is validation' to 'validation is iteration,' biological R&D AI is redefining the boundaries and pace of innovation in life sciences.