The bioeconomy is transitioning from costly trial-and-error "artisanal workshops" to an era of engineered "on-demand creation." On November 8, 2025, at the "2025 Lilly China Science Day" forum in Zhangjiang Science City, pharmaceutical giant Lilly announced the official launch of its open innovation platform, the Shanghai Innovation Incubator. Molecular Minds, a leader in AI protein design, became the first company to sign on as an incubator tenant.

The Lilly Shanghai Innovation Incubator aims to connect Lilly's global expertise and advanced technology platforms with China's most promising innovators, offering selected companies access to top-tier lab space and global resources. Molecular Minds' inclusion as the inaugural tenant underscores its technological prowess and industry value, earning recognition from a top-tier global pharmaceutical firm. As Dr. Julie Gilmore, Lilly's Global Vice President and Head of Innovation Incubators & Catalyze360 Portfolio Management, noted: "A new drug typically takes over a decade to move from lab to patient. Today, with iterative innovation and deeper integration of technology into medicine, we can significantly shorten and refine this lengthy process." Molecular Minds' selection as Lilly's partner in accelerating this journey highlights its technical foundation.

Behind this technological leap stands Xu Jinbo, a global leader in AI protein design and founder of Molecular Minds. At the core of this high-profile collaboration is the company's flagship product—MoleculeOS, the industry's first fully functional AI-powered protein optimization and design platform. The platform underwent a major upgrade this year, signaling a shift from "predicting" existing proteins to "creating" novel ones on demand. Beyond matching or surpassing AlphaFold 3 in predicting complex structures for key industrial applications, the platform has achieved a breakthrough in industrial-scale, ultra-high-precision dynamic protein design. It integrates dozens of solutions addressing critical pain points in biopharma and biomanufacturing.

This is no mere algorithm upgrade or tool enhancement. Zooming out from individual technologies to the broader industry reveals Molecular Minds' ambition: building the foundational infrastructure for the future digitalization, intelligence, and scalability of the entire bioindustrial chain—the "new infrastructure" of the bioeconomy. Just as computing centers and fiber-optic networks underpin the digital economy, a platform transforming protein R&D from "artisanal workshops" to "smart factories" is becoming the critical foundation for the bioeconomy's explosive growth—the infrastructure of industrial engineering.

1. The "New Infrastructure" Imperative: The Bioeconomy Demands Its Own Foundation "The prerequisite for an industry's takeoff is its ability to industrialize," Xu Jinbo, founder and chief scientist of Molecular Minds, emphasized in an interview. Long viewed as an "experimental science," biology has relied heavily on trial-and-error and serendipitous discoveries, particularly in protein and enzyme R&D, where traditional paths are fraught with uncertainty. A new drug follows the brutal "double-ten rule"—a decade and $1 billion, with less than a 10% success rate. Enhancing an industrial enzyme's activity might require a PhD team screening thousands of mutants over years—a high-stakes gamble.

"This costly trial-and-error model hinders industry growth," Xu noted. High capital barriers and sunk risks limit the game to deep-pocketed giants, burying the ideas of smaller innovators. The turning point came with two converging variables: massive data generation and powerful AI algorithms. High-throughput sequencing ushered life sciences into the big-data era, while AI, particularly deep learning, excels at uncovering patterns humans miss. AlphaFold's success epitomized this convergence, proving AI can transform unstructured biological information into precise, computable models.

With AI breaking core technical bottlenecks, the next industrial need is a foundational platform to scale and standardize this 0-to-1 capability for countless participants. This defines the "bioeconomy's new infrastructure": a bridge from academic innovation to mass production in life sciences and synthetic biology, turning "on-demand functional protein creation" into replicable, scalable services.

Yet reality bites. Building such an "industrial middle office" is daunting. First, generic AI protein tech alone can't solve industry problems. Xu explained that industrial scenarios are riddled with unique constraints—antibody design requires antigen-binding specificity, while industrial enzymes demand catalytic efficiency under specific temperatures or pH levels. "A generic model might generate millions of molecules before finding a viable one, but experiments can't scale that high," Xu stressed. "To reduce experimental load, we must embed proprietary constraints into generative models, necessitating specialized modules."

Second, a "valley of death" separates AI design from industrial deployment. Will AI-designed molecules express correctly in vivo? Function as predicted? Are they stable and safe? These require a dry-wet lab loop for rapid validation. Thus, an effective "new infrastructure" demands not just top algorithms but deep industrial process understanding and wet-lab validation. "You can't grasp key industrial pain points without hands-on experience—this know-how is critical," Xu added.

The bioeconomy's "new infrastructure" isn’t about open-sourcing algorithms or model APIs. It’s a complex system integrating massive data, specialized algorithms, automated workflows, industrial know-how, and validation loops—precisely what Molecular Minds is building.

2. Molecular Minds' Solution: Crafting the Bioeconomy’s "AI-Native OS" Facing these challenges, Molecular Minds' approach is clear: use "one foundational model + one integrated platform" to transform costly, slow, fragmented protein R&D into an engineered, scientific, industrial-grade service. This service is delivered via the MoleculeOS platform—an "AI-native" operating system. Unlike traditional platforms that annotate old maps, AI-native platforms redraw the map entirely. "AI-native" means the entire architecture and workflow are redesigned for AI’s new paradigm, where previously impossible tasks (e.g., end-to-end design) become feasible, necessitating a full workflow overhaul.

MoleculeOS decomposes industrial protein design into five categories: prediction, discovery, optimization, design, and validation. Each is supported by world-leading algorithm modules,串联成符合生物领域研发及生产的科学化流程。同时，分子之心自主研发了全球首个集成序列、结构、功能与进化的多模态AI蛋白基础大模型NewOrigin（达尔文）。它不再像传统模型那样仅从一维的序列“文本”去理解蛋白质，而是从一个“四位一体”的全息视角去认知生命分子。

"Proteins are products of evolution, mapping from sequence to structure to function," a Molecular Minds team member explained. "Traditional sequence-to-function models miss critical parameters." NewOrigin’s revolutionary advantage lies in handling complex, low-data scenarios and even enabling "zero-shot design," providing an AI cognition that deeply understands life’s principles for all platform applications.

Critically, Molecular Minds isn’t building a single technical peak but a platform integrating algorithms, data, industrial loops, and ecosystems. This solution’s foundation stems from Xu Jinbo’s two decades in AI protein research and the team’s accumulated data and tacit know-how from real-world projects—unattainable through paper replication or open-source models.

The company’s dry-wet lab and partner collaborations create an agile "design-validate-feedback-iterate" loop. This high-speed cycle accelerates technical iteration and sharpens industrial insights.

3. Industrial-Grade Thresholds and Winning Confidence: Solving Industry Problems with AI Platforms Molecular Minds’ innovation depth lies not just in algorithmic breakthroughs but in platformized engineering and industrial细节, delivering systemic solutions for rapid industry advancement. Key innovations include:

First, transcending static prediction for industrial-scale dynamic design. Proteins function dynamically, not as static snapshots. Traditional methods and most AI models rely on static structures, limiting enzyme design for catalytic reactions. "Without accurate dynamic structure prediction, dynamic design is baseless," the team noted. Molecular Minds uniquely merges AI and first-principles physics to achieve ultra-precise molecular dynamic prediction and design, overcoming AI’s small-data and knowledge-blind-spot weaknesses while simulating real chemical reactions. Its molecular simulations boast industry-leading precision and trillion-fold efficiency gains, reaching industrial-grade utility.

Second, superior complex-structure prediction. Protein functions often involve molecular complexes, making their prediction critical for applications. In antibody-antigen and enzyme-substrate scenarios, MoleculeOS matches or surpasses AlphaFold 3’s complex-structure precision, with better physical properties for reliable drug design and modification. "With precise complex structures, we need far fewer candidate molecules to find optimal compounds," Xu said, slashing screening from hundreds to dozens, saving time and costs.

Theory meets practice. Molecular Minds’ value lies not in "prediction" but in "delivering" industry-transforming results, venturing into R&D no-go zones where AI turns "impossible" into "achievable." Three examples:

- **Conditionally Activated Antibodies**: For pH-sensitive antibodies where high-throughput screening fails, MoleculeOS pinpointed <10 candidates in two months via two dry-wet loops, exhibiting strong pH sensitivity to potentially slash drug development timelines. - **Resurrecting a Drug**: A fusion protein drug nearly abandoned due to low expression and aggregation was saved by generative AI, boosting expression 400x with >95% monomers in three months, reviving a promising cancer therapy. - **"Super Industrial Enzyme"**: Partnering with biosynthetic leader Cathay Biotech, Molecular Minds tackled an enzyme with no crystal structure and complex catalysis, boosting efficiency 5x in six months without extensive client data.

Third, empowering biologists as "ability amplifiers." The biggest AI-biology integration challenge is the "knowledge gap." MoleculeOS addresses this by封装复杂算法为 "one-click" automated workflows, letting biologists use cutting-edge AI without expertise. Future conversational interfaces will allow natural-language commands like "design a heat-resistant enzyme."

"We’re not turning biologists into AI experts but making AI a super-tool for all," the team said. This usability dismantles barriers between biologists and AI, democratizing innovation.

4. Future Vision: Reshaping Industry Paradigms—AI Is the Key, Not the Endpoint AI’s role in biology is now a global consensus. The U.S. treats biotechnology as a national security priority, while Nobel laureate David Baker’s Xaira Therapeutics attracted over $1B for AI drug R&D. China’s "AI Plus" strategy similarly empowers cross-sector AI integration.

Like GPUs unleashed the deep-learning revolution, protein design capabilities are today’s "bottle opener" for bioindustrial value. Yet Molecular Minds’ impact transcends tools—it’s redefining the industry’s production model via an "AI-native OS" for the bioeconomy.

Historically, biotech followed a "discovery" paradigm—"what nature offers, we use"—with R&D as a high-risk, capital-intensive "gamble" under the double-ten rule. AI-native platforms flip this to a "creation" paradigm: "what we need, we design." This shift brings:

- **High Certainty**: AI’s精准计算 turns "needle-in-haystack" searches into targeted validation of high-potential molecules. - **Intelligence-Over-Capital**: Small teams leverage platforms to test ideas affordably,打破巨头垄断. - **Scalability**: Standardized, engineered R&D processes make biotech innovation scalable—a transformation more valuable than any single algorithm.

As Ming-dynasty scientist Song Yingxing wrote in *The Exploitation of the Works of Nature* (*Tiangong Kaiwu*), the core is harnessing nature’s principles for human benefit. Today’s AI protein design is the digital-era *Tiangong Kaiwu*—using AI to decipher life’s "divine craftsmanship" and create "super proteins" from scratch. Xu Jinbo and Molecular Minds aim to master the "heart" of molecules, wielding this wisdom to shape the future. This isn’t just corporate ambition—it’s a nation’s bid for leadership in core future technologies.

Whoever masters high-quality protein design and智能制造 engines will hold the reins to reshape the trillion-dollar bioindustrial chain.

(Cover image generated by AI)