字节Seed团队 投稿 量子位 | 公众号 QbitAI AI蛋白设计进入新阶段！ 最近，字节跳动Seed团队多模态生物分子结构大模型 （Protenix） 项目组提出了一种可扩展的蛋白设计方法，叫做 PXDesign 。 要知道，蛋白设计一直是个成功率很低的任务，即便是DeepMind推出的AlphaProteo，凭借其AlphaFold系列模型，在相同靶点上的成功率 也仅为 9%-33%。 此外，Protenix团队还推出了 公开免费的binder在线设计服务 ，让科学家无需自建复杂流程，就能直接调用这一能力，加速科研探索。 背景与意义 蛋白质是生命活动的基石。2024年诺贝尔化学奖一半授予 David Baker （计算蛋白设计） ，另一半联合授予 Demis Hassabis 与 John Jumper （蛋白结构预测） 。 在实际测试中，PXDesign展现出极高的效率， 24小时内即可生成数百个高质量的候选蛋白 ，生成效率较业界主流方法提升约10倍，并在多 个靶点上实现了 20%–73%的湿实验成功率 ，达到了当前领域的领先水平。 生成：快速高效 这也凸显了科学家关注的挑战： 不仅要"预测结构"， ...

撰文丨王聪 编辑丨王多鱼 排版丨水成文 Notch 信号通路 是一种进化上最保守的信号通路之一，在免疫细胞、神经元、血管内皮细胞、心肌细胞以及其他细胞谱系中是关键的发育命运决定因子，尤其是 T 细胞 的 发育和功能。 在体外从祖细胞/干细胞分化生成许多细胞类型都需要激活 Notch 信号通路。然而，在实验室中模拟这种高度机械性、依赖接触的信号通路，一直是个巨大的挑 战。 2025 年 8 月 1 日，哈佛大学医学院/波士顿儿童医院 George Daley 团队联合诺奖得主、蛋白质设计先驱 David Baker 教授，在国际顶尖学术期刊 Cell 上发表 了题为 ： Design of soluble Notch agonists that drive T cell development and boost immunity 的研究论文。 该研究利用 David Baker 教授团队开发的 AI 蛋白 设计工具 Rosetta ，成功设计生成了全球首个 可溶性 Notch 激动剂 ，这种可溶性蛋白能够在悬浮培养中模拟 Notch 信号通路的激活，实现了 T 细胞高效分化，并显著增强了 T 细胞功能及其抗肿 ...

Core Insights - The AI protein design company "MoleculeOS," founded by Xu Jinbo, showcased significant advancements at the WAIC 2025, marking a breakthrough in the AI protein design field [1][3] Group 1: Technology and Innovation - MoleculeOS is an industry-grade AI protein infrastructure platform that integrates the world's first multimodal AI protein foundation model, NewOrigin (Darwin), along with over ten leading AI protein prediction, optimization, and design technologies [1] - The platform demonstrates superior performance, surpassing AlphaFold 3 in complex structure prediction, achieving not only comparable accuracy but also improved physical properties for better antigen-antibody and enzyme-substrate complex predictions [3] - MoleculeOS has significantly enhanced molecular simulation precision and efficiency, achieving a million-fold increase in efficiency, reaching industrial-grade levels [3] Group 2: Industry Applications - The platform has been optimized to meet industry demands, offering automated workflows for drug design, enzyme design, and various other applications, including antibody design and enzyme stability [4] - MoleculeOS has been validated in multiple industrial projects, addressing real-world needs in innovative drug design and synthetic biology, allowing for customized protein design with a single click [4] Group 3: Accessibility and Impact - MoleculeOS features a conversational AI agent, enabling biologists without an AI background to design high-value molecules quickly and accurately, thus lowering the technical barrier [6] - The traditional methods of protein design are time-consuming and have low success rates, but AI integration offers a transformative approach, significantly improving research efficiency and success rates in drug development [6] - MoleculeOS aims to empower biologists by freeing them from tedious laboratory tasks, allowing them to focus on strategic design and judgment, ultimately leading to safer, more effective drugs and lower-cost bioproducts [6]

Core Viewpoint - Synthetic biology is poised to reshape the world through innovations such as novel antibody drugs, mRNA vaccines, and DNA data storage, with DNA synthesis as the foundational technology [1] Industry Status: Supply and Demand "Time Lag" - The high-throughput DNA synthesis sector in China is experiencing a "time lag" between supply and demand, with various applications yet to fully materialize [4] Challenges and Opportunities on the Marathon Track - Increased capital influx has led to intense competition focused on capacity and pricing, hindering the establishment of true technological barriers and sustainable business models [5] - The company aims to be a "marathon runner" in the industry, focusing on long-term value creation and resilience through economic cycles [5] Performance Priority and Cost Optimization - The company prioritizes performance as the cornerstone of technological value, ensuring high-quality synthesis before pursuing cost reductions through innovation and process optimization [6] Robust Operational System: Building a Value-Driven Moat - Long-term development is rooted in core value and robust operational capabilities, with a focus on customer value and a resilient operational system [7] - The company controls the entire supply chain from raw materials to delivery, enhancing risk resistance and ensuring stable delivery [7] Supply Side: Industry Outlook Attracting Capital - The high-throughput DNA synthesis market has seen over twelve companies emerge, with intense price competition and rapid cost declines [8] - The market capacity is projected to be between 100-150 million RMB by 2025, with potential demand growth offset by rapid price declines [8] Core Technology Deep Self-Research - The company has developed a high-throughput synthesis platform from scratch, continuously optimizing key processes to ensure quality while optimizing costs [9] Ignite 3.0: Reducing Burden and Accelerating Innovation - The company has launched the Ignite 3.0 platform, designed to address core industry pain points and enhance the development of synthetic biology [10] - Ignite 3.0 integrates high throughput, quality, and short cycles, supporting oligo pools of varying sizes and achieving a low error rate of below 0.2% [11] Performance Data Evidence - The platform can synthesize 680,000 independent points in a single run, with a synthesis length of up to 200nt and a coverage rate exceeding 99.9% [12] - The uniformity of the synthesized products is demonstrated with a 95/5 percentile ratio of 1.82, ensuring equal sampling in downstream screening [14] Future Prospects: Open Experience - Ignite 3.0 is not just a synthesizer but a validated "full-process solution," with high-quality oligo pools already showing strong competitiveness in various fields [19] - The company invites research users to apply for testing the Ignite 3.0 platform, aiming to expand into high-throughput gene construction [19]

Core Insights - The article discusses the upgrade of Tushen Zhihuo's protein design platform to ProteinNova, which now features AI-driven full-process protein design capabilities [1][5] - The integration of dry and wet lab processes allows for a closed-loop system in AI protein design, enhancing the efficiency of scientific research and product development [2][5] Group 1: AI-Driven Protein Design - The upgraded platform showcases a complete closed-loop process from AI-generated design to experimental validation, emphasizing the importance of real-world testing in scientific research [2] - The platform's capabilities include task management, data tracking, and feedback integration, enabling users to create personalized iterative tasks for rapid prototype validation [2][5] Group 2: Optimization of Decision-Making - The system's reasoning process has been streamlined to enhance clarity and user experience, allowing users to quickly grasp core insights without being distracted by unnecessary steps [3] - A complete logical chain is maintained for users who wish to delve deeper into the AI's reasoning, ensuring transparency while keeping the process efficient [3] Group 3: Reporting Enhancements - The introduction of structured report templates allows AI to extract key information such as design logic and improvement suggestions, improving report readability and usability [4] - The language of the reports has been optimized to balance scientific rigor with accessibility for researchers [4] Group 4: Future Implications - The launch of the dry-wet iteration system marks a significant advancement in AI's role in scientific research, enabling rapid feedback and self-optimization [5] - The platform aims to support a wider range of disciplines and experimental types, moving towards the realization of "automated science" as a foundational tool for research teams [5] Group 5: Company Background - Tushen Zhihuo focuses on accelerating scientific breakthroughs and product innovation through super-intelligent platforms, significantly reducing the cycle time for new product design [6] - The core team comprises experts from prestigious institutions, bringing extensive experience in AI research and commercialization [6]