AI Drug Discovery Platforms: Breaking the "Double Ten" Law of New Drug Development

In the pharmaceutical industry, there is a long-standing "Double Ten Law": a new drug from R&D to market is widely recognized to take an average of ten years and cost over one billion US dollars. Behind this law lies the industry's long-standing challenges of high investment, long timelines, and low success rates.

In the traditional drug R&D model, researchers must screen hit compounds from massive compound libraries, followed by multiple rounds of structural optimization, animal studies, and clinical validation. Failure at any stage means years of time and tens of millions of dollars in investment lost. The intervention of AI is now changing these rules of the game that have governed the industry for over half a century.

I. AI Drug Discovery: A Market on the Verge of Explosive Growth

Market Growth

According to iiMedia Research data, the global AI drug discovery market grew from $790 million in 2021 to $2.41 billion in 2025. Huachuang Securities research reports indicate that China's AI drug discovery market reached 560 million RMB in 2024 and exceeded 600 million RMB in 2025. Other research reports suggest the global AI drug discovery market was approximately $2.49 billion in 2025 and is expected to surpass $46 billion by 2035.

The rapid market growth stems from a core driving force: the persistent efficiency challenges in the pharmaceutical industry are forcing technological upgrades. The traditional drug discovery process typically takes 3 to 6 years, and AI platforms are effectively compressing this timeline.

II. How AI Is Transforming Drug R&D

AI Discovery Workflow

AI's empowerment of drug R&D spans the entire chain from target discovery to clinical trials.

Target Discovery and Validation. AI rapidly identifies potential disease-related targets by learning from massive literature, genomic data, and protein structure information. Traditional methods rely on researchers manually screening literature and experimental data, which is time-consuming and limited in scope. AI can integrate massive multi-omics and literature data within days, assisting researchers in quickly narrowing down potential targets and compressing the early target discovery research cycle from months to days.

Molecular Generation and Virtual Screening. Generative AI can design entirely new molecules with specific pharmacological properties from scratch, breaking through the limitations of traditional "known compound library" approaches. AI-driven virtual screening can rapidly evaluate the binding activity of millions of molecules on standard computing clusters. Compared to traditional virtual screening, AI approaches improve positive predictive accuracy by 30%–50%, significantly reducing downstream experimental validation workload.

ADMET Prediction. A drug's Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) profile is a core component of preclinical evaluation. AI models trained on massive experimental datasets have achieved high accuracy in predicting properties such as absorption and solubility. At the molecular design stage, AI can pre-eliminate over 80% of high-risk molecules, effectively reducing late-stage failure rates.

Clinical Trial Optimization. AI can assist in patient recruitment, protocol design, and data monitoring, improving clinical trial efficiency.

III. AI Protein Design Platforms: The "Foundational Engine" of Drug R&D

Protein Design Engine

In the small molecule drug space, AI empowers the screening and design of small-molecule compounds. In the biologic drug space, however, proteins are the drugs themselves—antibodies, enzymes, peptides, and fusion proteins are all proteins. The structural and functional complexity of proteins makes them one of the areas with the greatest potential for AI empowerment.

Matwings Technology's core technological driver is its leading AI protein design platform,   AIACCLBIO®. The platform is pre-trained on a proprietary protein dataset containing nearly 9 billion sequences—a dataset that not only covers conventional biological protein sequences but also integrates specialized functional sequences from extreme environments such as volcanoes and deep seas, sourced through the "Abyss Plan." Among these, nearly 500 million sequences are annotated with functional labels including temperature and pH. AIACCLBIO® features two core capabilities—"AI-directed evolution" and "AI enzyme mining"—combined with few-shot learning algorithms and a dry-wet iteration model to achieve end-to-end "sequence-to-function" prediction.

On April 24, 2026, Matwings Technology launched its conversational protein R&D intelligent agent, MatwingsVenus™ （XIAOWU™）. This agent-centric platform provides a one-stop protein R&D solution, supporting retrieval from billions of real labeled protein data points, integrating over 200 protein design tools, more than 50 platform-certified experts, and over 30 domain-specific fine-tuned Skills. Users simply input task objectives via natural language, and the system automatically decomposes tasks and orchestrates corresponding design, prediction, analysis, and screening capabilities.

The MatwingsVenus™ （晓鹜™） platform bridges the gap between cloud-based design and physical experimentation, establishing a "conversational dry-wet closed loop." After the Agent completes its design, the platform's proprietary communication mechanism automatically routes results to plasmid ordering and experimental scheduling workflows, seamlessly coordinating subsequent experimental tasks and directing robots to complete sample preparation, protein purification, and functional testing. Experimental results flow back to the next round of AI design, forming an iterative closed loop of computation-driven wet-lab experimentation and wet-lab data-informed computation.

In terms of industrial capabilities, according to Matwings Technology's official information, the platform has successfully delivered over 30 protein design industrialization projects, with more than 40 additional projects in the pipeline. From market, literature, and patent research, through molecular design and small-scale validation, to pilot process optimization and production scale-up, the platform has established a comprehensive capability system covering the entire drug R&D value chain.

IV. Commercialization Milestones in AI Drug Discovery

The value of AI in drug R&D has moved beyond proof-of-concept to commercial realization.

The collaboration between Matwings Technology and GeneScience Pharmaceuticals is a landmark case in AI-driven process development. Leveraging a protein engineering general foundation model combined with limited wet-lab closed-loop iterative validation, the R&D team transformed an ordinary non-alkali-resistant single-domain antibody—achieving a 4-fold improvement in alkali resistance and doubling its lifespan—in just 4 months, successfully scaling the process to 5,000-liter production. This product is the world's first protein product designed by a foundation model that has entered 5,000-liter scale-up production and commercial application, saving the partner enterprise tens of millions of RMB annually in protein raw material costs.

In a de novo design project targeting an immune-regulatory receptor, Matwings Technology, leveraging its independently developed MatwingsVenus™ （xiaowu™） platform, successfully obtained dozens of novel binder molecules with demonstrated in vitro cell-blocking activity. The target was novel with limited precedent, lacked reference drug molecules, and featured a surface dominated by polar regions without typical high-druggability binding hotspots—combined with a natural ligand already exhibiting nM-level affinity, which significantly compounded the design challenge. Leveraging the platform, the Agent autonomously completed the full computational workflow—including scaffold screening, interface design, sequence optimization, and druggability prediction. Samples prepared via the automated experimental platform demonstrated outstanding performance in in vitro cellular activity assays, with dozens of molecules exhibiting clear cell-blocking activity.

Matwings Technology's AI protein design capabilities have received multiple industry recognitions. In July 2025, Matwings Technology was named to the Ministry of Industry and Information Technology's first batch of "Typical Application Cases of Artificial Intelligence in Biomanufacturing," achieving an "Excellent" rating. It was also listed on the 2025 World Artificial Intelligence Conference (WAIC) SAIL Award TOP 30. In June 2026, Matwings Technology was selected for 36Kr's "2026 Most Valuable Growth Enterprises 100" in the Artificial Intelligence and Foundation Models track.

V. Challenges and the Future

Despite the rapid development of AI drug discovery, the industry still faces multiple challenges.

Data Quality and Data Scarcity. The effectiveness of AI models heavily depends on the quality and scale of training data. Drug R&D data is extremely costly to obtain, and a large volume of data remains siloed across different institutions—the data silo problem remains prominent.

Model Generalization. A model trained on a specific target and molecular type cannot be easily transferred to new targets or new molecular types. Cross-scenario generalization capability still requires continuous improvement.

Experimental Validation Closed Loop. AI-generated designs ultimately require wet-lab validation. How to efficiently bridge computational design and experimental validation to form a rapid iteration closed loop is a core challenge that AI drug discovery platforms must address.

Regulatory Compliance. How AI-designed drug molecules will pass regulatory review, and how AI-assisted clinical trial data will be accepted by regulatory agencies—the relevant regulatory frameworks are still under development.

Nevertheless, industry consensus is forming: AI spending in drug R&D is projected to grow from $4 billion in 2025 to $20 billion by 2030. When AI can design drug molecules as precisely as designing chips, it is expected to compress the drug discovery phase from 3–6 years to 6–12 months, significantly reducing early-stage R&D costs. Drug R&D will no longer be the exclusive domain of large pharmaceutical companies—and AI drug discovery platforms stand as the core engine of this transformation.