Absci Corp.

| 从数年缩短到数月，AI正在彻底改写新药研发的游戏规则核心摘要： 人工智能技术正以前所未有的速度渗透到药物研发全链条，从靶点发现、分子设计到临床试验优化，AI将传统药物研发周期从10-15年缩短至2-3年，成功率从不足10%提升至30%以上，成为全球制药行业的技术革命。   01 降维打击：AI如何颠覆传统研发   传统药物研发被称为"双十定律"——耗时10年、耗资10亿美元，成功率却不足10%。这一魔咒正在被AI技术打破。   深度学习算法能够分析海量的生物医学数据，包括基因组学、蛋白质结构、临床试验数据等，快速识别潜在的药物靶点。与传统方法相比，AI筛选化合物的速度提升1000倍以上，成本降低90%。   "AI不是简单地替代人工，而是在做人类无法完成的事情，"一位资深药物研发专家表示，"它能在数百万种化合物中快速找到最有希望的候选药物，这是传统方法无法想象的。"   02 重大突破：AI驱动的新药问世   近期，多个由AI驱动研发的新药进入临床阶段，标志着这一技术从概念走向现实。   英矽智能（Insilico Medicine）利用AI平台发现的特发性肺纤维化治疗药物，从靶点发现到临床候选化合物确定仅用时18个月，创下行业纪录。该药物目前已进入二期临床试验。   Recursion Pharmaceuticals利用AI技术发现的神经纤维瘤病治疗药物，获得FDA突破性疗法认定，预计2025年提交新药申请。   BenevolentAI开发的特应性皮炎治疗药物，在二期临床试验中显示出显著疗效，预计2024年进入三期临床。   03 资本狂热：千亿市场正在形成   AI药物研发赛道正成为资本追逐的新风口。2024年前三季度，全球AI制药领域融资额超过50亿美元，同比增长120%。   巨头布局：  辉瑞投资5亿美元与AI制药公司合作开发肿瘤药物  默沙东与Absci达成6.1亿美元合作，利用AI设计新型抗体药物  罗氏收购Flatiron Health，加强AI在肿瘤数据分析中的应用   中国企业崛起：  晶泰科技完成4亿美元D轮融资，估值超30亿美元  未知君生物完成1亿美元B轮融资，专注微生物组药物研发  剂泰医药获数千万美元投资，AI驱动制剂开发  泓博医药布局AI制药不断副能客户新药研发，24年底服务客户40家，新药项目超过80成都先导于腾讯合作，研究分子股价跃迁算法 04 技术前沿：多模态AI的突破   新一代AI药物研发平台正从单一模态向多模态融合发展。   AlphaFold 3的发布标志着蛋白质结构预测进入新纪元，能够准确预测蛋白质与配体、核酸的相互作用，为药物设计提供更精准的蓝图。   生成式AI在药物分子设计中的应用取得突破，能够根据特定靶点"凭空"生成具有理想性质的化合物结构，大大扩展了化学空间。   强化学习算法在优化临床试验方案方面表现出色，能够根据实时数据动态调整试验设计，提高试验效率和成功率。   05 临床价值：从实验室到病床   AI药物研发的真正价值在于为患者带来更好的治疗选择。   个性化医疗：AI能够分析患者的基因组数据，预测药物反应和副作用，实现真正的精准用药。   罕见病突破：传统研发模式下，罕见病药物因市场小而难以获得投资。AI大幅降低研发成本，使得针对罕见病的药物开发变得经济可行。   老药新用：AI能够发现已有药物的新适应症，让"沉睡"的药物重新焕发价值。例如，AI发现二甲双胍可能具有抗衰老作用，相关研究正在进行中。   06 挑战与机遇：前路依然漫长   尽管AI药物研发前景广阔，但仍面临诸多挑战。   数据质量：AI模型的高度依赖高质量的训练数据，而医疗数据的标准化和可获得性仍是瓶颈。   监管适应：监管机构需要建立新的审评标准和方法，以适应AI驱动的研发模式。   人才短缺：既懂AI又懂药物研发的复合型人才极度稀缺，成为行业发展的制约因素。   伦理考量：AI算法的透明度和可解释性至关重要，特别是在涉及生命健康的领域。   07 未来展望：智能医疗新纪元   业内专家预测，到2030年，50%以上的新药研发将深度整合AI技术，全球AI制药市场规模将超过1000亿美元。"我们正站在医疗健康产业的历史转折点，"一位行业领袖表示，"AI不仅改变着药物研发的方式，更在重新定义人类对抗疾病的能力边界。"从靶点发现到临床用药，从罕见病到常见病，AI正在为每一个疾病领域带来新的希望。这场技术革命的意义，不仅在于研发效率的提升，更在于为无数患者带来生命的曙光。

The Japanese pharma giant Takeda has signed a second research deal with the AI specialist Nabla Bio, expanding the scope of a collaboration that started three years ago. The two companies announced the deal Tuesday, with Takeda agreeing to pay double-digit millions in upfront and research cost payments, along with possible milestones of more than $1 billion. Nabla’s leaders said there was mutual interest in expanding their work after seeing successes with Nabla’s AI models for generating and designing proteins. There has been a flurry of small, AI-focused deals in the last few weeks as big drugmakers explore a range of AI approaches in R&D. Within the past month, Merck expanded a partnership with Variational AI on small molecules, AstraZeneca signed deals with Algen Biotechnologies and Turbine , and Sanofi announced a three-year deal to use BenchSci’s AI platform on disease biology. Nabla, a Cambridge, MA-based biotech that spun out of George Church’s lab in 2020, has focused on pharma partnerships rather than building its own pipeline. The 20-person company raised a $26 million Series A last year , and plans to grow to 25 to 30 employees over the next couple of months, CEO Surge Biswas said. It has also partnered with AstraZeneca and Bristol Myers Squibb. “One of the things that has been really exciting to us is the broad scope of impact,” Nabla’s president and co-founder Frances Anastassacos said in an interview. “We want to get the technology to influence as many programs as possible and as quickly as possible.” Hello, Sheri Morales! You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Hello, Sheri Morales! You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Hello, Sheri Morales! You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Hello, Sheri Morales! You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Hello, Sheri Morales! You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Hello, Sheri Morales! You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Hello, Sheri Morales! You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Hello, Sheri Morales. You’re an essential part of the biopharma world. Share your voice on this subscriber Pulse Survey to help shape our understanding of the industry. Results will be anonymous. Since signing that first Takeda deal in 2022, Nabla and the broader AI bio field have made dramatic progress with AI models to generate and design proteins. Last year, Nabla debuted its AI model called JAM , which has made antibodies from scratch that bind to some hard-to-drug targets. Biswas described it as “autocomplete for molecules.” Given context on a research problem, like the protein sequence and structure of the target, JAM “can autocomplete the biologic that binds to that target at a specific epitope location.” Today, Biswas said they are seeing double-digit percent success rates in getting hits with JAM that bind fairly strongly to a target — often in the double-digit nanomolar and sometimes in the single-digit nanomolar range. “That is one dimension that needs to improve,” Biswas said, adding they are “pushing hard” on getting down to picomolar binding affinities. The tighter drugs bind to their target, the better — it can lead to higher potency and less potential for off-target side effects. JAM is a key part of the new deal, in which Nabla will work on a range of Takeda research projects. Focus areas include de novo biologics for antibodies and multispecifics. The work will include targets that can’t be hit using conventional methods, but also targets that “could be accessed with traditional discovery approaches but are very inefficient in that way,” Biswas said. “With the same resources that it would ordinarily take to process a single target, we can process five to 10 in parallel,” he said. Nabla will also work on optimizing some existing biologics that have severe developability issues, in what Biswas called rescuing programs and “bringing it out of the graveyard.” Across its partnerships, Nabla has two programs “getting very close to IND-enabling studies,” Biswas said. Entering the clinic “in the next year or two is not too crazy,” he said, although that decision will ultimately be made by the partnering pharma companies. That may make Nabla potentially the first company to bring a de novo protein into the clinic. Other AI-focused biotechs like Generate:Biomedicines and Absci are advancing clinical-stage programs where AI played a role in their development, but that aren’t de novo efforts. A key unknown for AI-made proteins is their immunogenicity in the human body, Biswas said. Oftentimes, anti-drug antibody responses appear in the clinic that are hard to predict with preclinical models or animal testing. The immunogenicity question “is something the industry overall just doesn’t have a great grasp on yet, and is sort of the big question mark with computationally designed antibodies, because we have yet to see one of these really get into the clinic and see how they fare in humans,” Biswas said. “No one’s going to feel totally comfortable until we see these in a human.” But Biswas said he’s encouraged by the preclinical data and in Nabla’s efforts to make its de novo proteins “indistinguishable from antibody sequences you would pull out of a human.”

iStock, Robert Way AstraZeneca has invested heavily in AI, primarily through collaborations, including an up to $5.3 billion partnership with China’s CSPC Pharmaceutical in June. AstraZeneca is adding to its roster of AI-focused partners, this time looking to California-based Algen Biotechnologies to advance novel therapies in immunology. The companies did not provide an exact financial breakdown of the deal, announced Monday, only revealing that the total consideration could reach up to $555 million, including upfront and near-term payments, plus development, regulatory and commercial milestones. The partners likewise did not reveal what diseases they will prioritize, noting only that they will work on “therapies against a defined set of targets” in immunology. Algen spun out of Nobel Laureate Jennifer Doudna’s lab at the University of California, Berkley to work on its AlgenBrain platform —the centerpiece of Monday’s deal—which, according to Monday’s release , captures “billions” of RNA changes in humans to elucidate “causal links between gene regulation and disease progression.” The biotech then uses those data, along with machine learning, to identify novel therapeutic targets that “may reverse disease processes.” Doudna is an adviser to Algen. “We are using AI and Machine Learning to enhance the discovery of new targets,” Jim Weatherall, AstraZeneca’s chief data scientist of Biopharmaceuticals R&D, said in a statement, noting that leveraging such technologies “will help to create more effective medicines.” “Algen’s platform offers a powerful approach to achieving this,” he added. Aside from Algen, AstraZeneca in recent years has heavily invested in its AI capabilities, largely through partnerships. “We know that we’re not an AI company but that we have to partner,” Arun Krishna, the pharma’s head of U.S. oncology for lung cancer, told BioSpace in June. That month, for instance, AstraZeneca fronted $110 million to link up with China’s CSPC Pharmaceutical to gain access to its AI capabilities for novel oral therapies targeting several chronic diseases. The pharma also put up to $1.62 billion on the line in development milestones, plus $3.6 billion in sales-based payments. Two months earlier, in April, AstraZeneca signed a three-way agreement with Pathos AI and Tempus AI, in a deal that will see all parties work together to “build a multimodal foundation model in oncology,” yielding new biological and clinical insights to guide drug development. The financial details of this alliance remain murky, though AstraZeneca has pledged $200 million in data licensing payments. In 2023, AstraZeneca also inked two AI collaborations. One came in September that year with Verge Genomics, involving more than $880 million in commitments, targeting rare neurodegenerative and neuromuscular conditions. The other was in December, with Absci , focused on identifying a cancer candidate. Payments to Absci could hit $247 million.