Anthropic 希望研发自己的药物
内容总结：
Anthropic发布“科学家AI工作台”，宣布自研新药布局前沿科学领域
本周早些时候，在“AI for Science”简报会上，人工智能公司Anthropic正式推出名为Claude Science的新产品。这是一款面向科学家的“AI工作台”，旨在将零散的工具与数据集整合至统一环境，并支持生成图表和可视化内容。Anthropic凭借其广受欢迎的编程工具和强大的AI模型已在行业中占据主导地位，此次发布围绕AI“大幅加速科学发现和医疗干预研发进程”的潜力展开，并列举了多家已在使用Claude的生物技术与制药客户。
Anthropic计划自研药物，加入AI制药竞赛
Anthropic更进一步，宣布将自行研发药物。生命科学负责人Eric Kauderer-Abrams表示，公司将专注于为“被忽视”的疾病寻找治疗方案。当前，AI公司争相拓展科学和制药客户群——OpenAI、亚马逊、谷歌等均已推出各自的生命科学工具和平台。但Anthropic此举是主要前沿AI公司中最为直接的公开尝试之一，亲自下场研发药物。这使其处于一个特殊位置：既向其他潜在竞争药企销售软件，又亲自参与竞争。Anthropic加入的这场更广泛的竞赛中，包括以AI为先导的Insilico Medicine、谷歌DeepMind衍生的Isomorphic Labs、多家生物技术初创公司，以及正在自建或购买AI工具的大型制药企业。
然而，Anthropic迄今未就药物开发领域的具体目标提供详细说明。在发布会上，Kauderer-Abrams并未透露若发现候选药物后公司将如何行动。Anthropic也未回应The Verge的进一步询问，包括计划优先针对哪些疾病、是否会与其他公司合作进行实验室工作、动物试验、临床试验或生产等细节。
专家：AI已渗透药物研发全流程，但离患者还很远
剑桥大学教授兼AI生物技术初创公司CardiaTec联合创始人Namshik Han向The Verge表示，Anthropic计划的不确定性正反映了AI制药热潮本身的不确定性。他指出，“AI药物发现”是一个宽泛的术语，AI已被应用于“药物发现的每一个阶段”，从发现新化合物、优化化合物，到支持研究、数据分析、临床试验乃至生产。伦敦大学学院药物发现教授Matthew Todd也认为AI已全面渗透药物研发领域，并将其称为一个“包罗万象的术语”。
AI无疑正在改变药物开发。Han指出，阿斯利康、诺和诺德、葛兰素史克等制药巨头已推出多项举措，AI已能帮助生成可能的药物构想，例如推荐可能与已知疾病相关受体或现有药物靶点相互作用的新分子。Todd表示，AI在加速研究、帮助“路测”新药构想方面极为有用。鉴于Anthropic在前沿模型方面的积累，该公司很可能会利用生成式AI在广阔的化学生物可能性中搜索，帮助研究人员建立难以或缓慢发现的关联，进而提出新药构想、识别新疾病靶点或为现有药物寻找新用途。
但即便如此，距离AI设计的药物真正惠及患者仍有很长的路。Todd表示，从AI设计的新药到获得监管机构批准用于人体，“还差得很远”。他补充说，药物发现过程无法自主运行，全程需要人类投入和监督。Todd和Han还共同指出，缺乏公开的高质量实验数据——例如各种化学物质在体内的行为表现——也可能拖慢药物研发进程，即使在研究较为充分的生物学领域，我们对事物运作机制的理解仍存在巨大空白。
AI模型“远未让实验变得多余”
牛津大学结构化学生物学教授兼牛津药物发现中心蛋白质晶体学主任Frank von Delft表示，人们对AI模型的进步感到兴奋是合理的，但“它们远未让实验变得多余”。候选药物仍须在现实世界中测试其有效性、毒性，以及是否具备作为药物制备、储存和安全递送的实际特性。所有这些都需要熟练的劳动力、大量资金和时间，尤其是涉及人体的临床工作——这正是许多有前景的候选药物夭折的节点。Von Delft指出，如果Anthropic想开发药物，“就必须在实验上投入巨资”。
Anthropic似乎有意为此付出努力。过去一年中，该公司积极招聘生物学家并建设自己的湿实验室，截至发稿时仍有多个生命科学岗位开放招聘。Han表示Anthropic正在该领域“积极招兵买马”，多位学术同行已被该公司接触。他认为Anthropic已成功从大型制药企业和知名学术机构挖走数名人才。
考虑到所有这些复杂性，无论Anthropic选择攻克哪种疾病，获得回报都可能遥遥无期——至少需要十年左右的时间，因为新药通过临床试验通常需要漫长周期。Todd表示，“试验药物总是存在很大的滞后时间”，“证明某种东西安全需要时间”。目前，尚无AI设计的药物成功通过临床试验并获得FDA批准上市。部分AI开发的候选药物已进入临床试验，但很难判断AI在其中贡献了多少、在哪个环节被使用，或者这些候选药物是否优于传统药物。AI可以加速部分搜索过程，但药物仍需以传统方式证明自己：在现实世界中通过缓慢、有条理的实验来验证。
中文翻译：
在本周早些时候举行的“AI for Science简报会”上，Anthropic发布了Claude Science，这是一个面向科学家的新型AI工作台，它将分散的工具和数据集整合到一个环境中，并能够生成图表和可视化内容。Anthropic凭借其广受欢迎的编程工具和强大的AI模型已在行业内占据主导地位，此次发布围绕其所称的AI“能极大加速科学发现和医疗干预手段开发”的潜力展开，并列举了一长串已经在使用Claude的生物技术和制药客户。
Anthropic希望开发自己的药物
AI药物热潮距离惠及患者仍有很长的路要走。
Anthropic更进一步，表示将自行研发药物。生命科学负责人Eric Kauderer-Abrams表示，公司将专注于为“被忽视”的疾病发现治疗方法。
AI公司一直积极争取科学和制药领域的客户——OpenAI、亚马逊、谷歌等公司都拥有自己的生命科学工具和平台。但Anthropic的这一步棋，是主流前沿AI公司中最为直接地公开尝试自行研发药物的一次。这使得该公司处于一个不寻常的位置：它既向其他可能构成竞争的制药公司销售软件，又亲自下场。Anthropic加入了一场更广泛的竞争，参与者包括Insilico等AI优先的制药公司、谷歌DeepMind分拆出来的Isomorphic Labs、生物技术初创公司，以及正在自建或购买AI工具的大型制药公司。
对于在药物开发领域的具体目标，Anthropic提供的细节极少。在简报会上，Kauderer-Abrams并未说明如果发现任何有前景的候选药物，公司会怎么做。Anthropic没有回应The Verge希望了解更多细节的评论请求，包括计划首先针对哪些疾病，以及是否会在实验室工作、动物测试、临床试验或生产环节与其他公司合作。
AI被应用于“药物发现的每一个阶段”。
专家告诉The Verge，围绕Anthropic计划的不确定性，反映了AI药物热潮本身存在更广泛的不确定性。“AI药物发现”可以指代很多事物。剑桥大学教授、AI生物技术初创公司CardiaTec联合创始人Namshik Han解释说，这是一个“非常宽泛的术语”。他说，AI被应用于“药物发现的每一个阶段”，从发现新化合物并加以改进，到支持研究、数据分析、临床试验，甚至生产。他表示，每家大型制药公司都会以某种方式使用AI。伦敦大学学院药物发现教授Matthew Todd也赞同AI已渗透到药物发现和研究中的看法，鉴于其用途广泛，他称之为一个“总括性术语”。
AI无疑正在改变药物开发。Han提到了阿斯利康、诺和诺德和葛兰素史克等制药巨头的大量举措，并表示AI已经能够帮助生成可能的药物构想，例如建议新的分子，这些分子可以与体内已知与特定疾病相关或已作为现有药物靶点的细胞受体等部位相互作用。Todd表示，AI对于加速研究和帮助“路测”新药构想极其有用。鉴于Anthropic在前沿模型方面的工作，该公司可能会利用生成式AI来探索广阔的化学和生物学可能性，帮助研究人员建立那些否则难以或缓慢发现的联系，可能提出新的药物构想、识别新的疾病靶点或为现有药物找到新用途。
但这距离AI设计的药物惠及患者仍有很长的路。Todd表示，从AI设计的药物获得监管机构批准用于人体，该领域“还有很长的路要走”。他补充说，药物发现过程不会自主运行，全程都需要人类的输入和监督。Todd和Han都指出，缺乏公开的高质量实验数据（例如各种化学物质在体内的表现）也可能会拖慢药物开发工作，并强调即使是在研究充分的生物学领域，我们对事物运作方式的理解仍存在巨大空白。
AI模型“距离让实验变得可有可无还差得很远”。
AI无法解决药物发现过程中许多最耗时的环节。牛津大学结构化学生物学教授、牛津药物发现中心蛋白质晶体学负责人Frank von Delft表示，人们对推进AI模型感到兴奋是对的，但AI“距离让实验变得可有可无还差得很远”。候选药物仍必须在现实世界中测试其有效性、毒性，以及是否具备能够作为药物安全制备、储存和递送的实际特性。所有这些都需要熟练的工人、大量的资金和时间，尤其是在人体上的临床工作——这也是许多有希望的候选药物失败的节点。von Delft说，如果Anthropic想开发一种药物，它“将不得不在实验上投入大量资金”。
Anthropic或许愿意尝试。在过去一年里，该公司一直在积极招聘生物学家并建设自己的湿实验室。截至本文撰写时，它仍有几个正在招聘的生命科学职位。Han表示，Anthropic也在该领域“积极招聘”，并补充说他的几位学术同行已被该公司接洽过。Han没有点名，但他认为Anthropic已成功从大型制药公司和知名学术机构挖走了一些候选人。
考虑到所有这些复杂性，无论Anthropic选择哪种疾病，任何回报都可能遥遥无期——考虑到一种新药通过临床试验通常所需的时间，至少也是近十年的事。Todd说，药物测试“总是存在很大的滞后时间”。“通过实验证明某些东西是安全的，这需要时间。”目前还没有任何AI设计的药物通过临床试验并获得FDA批准上市。一些AI开发的候选药物已进入临床试验，但很难知道AI贡献了多少、在过程的哪个阶段被使用，或者这些候选药物是否优于传统药物。AI可以加速部分搜索过程，但药物仍然需要以老式方法来证明自己：在现实世界中通过缓慢、有条理的实验来进行。
英文来源：
At the event “The Briefing: AI for Science” earlier this week, Anthropic announced Claude Science, a new “AI workbench for scientists” that pulls fragmented tools and datasets into one environment, and generates figures and visuals. Anthropic, already dominating the industry with its popular coding tools and powerful AI models, framed the launch around what it says is AI’s potential to “dramatically accelerate the pace of scientific discovery and the development of healthcare interventions,” and touted a long list of biotech and pharma customers already using Claude.
Anthropic wants to develop its own drugs
The AI drug boom has a long way to go before reaching patients.
The AI drug boom has a long way to go before reaching patients.
Anthropic also went a step further, saying it would develop drugs of its own. Head of life sciences Eric Kauderer-Abrams said the company will focus on discovering treatments for “neglected” diseases.
AI companies have been eager to court science and pharma customers — OpenAI, Amazon, Google, and others have their own life sciences tools and platforms. But Anthropic’s planned move is one of the most direct public attempts by a major frontier AI company to actually develop drugs itself. It puts it in the unusual position of selling software to other, potentially competing drugmakers. Anthropic joins a broader race that includes AI-first drug companies like Insilico, Google DeepMind spinout Isomorphic Labs, biotech startups, and Big Pharma companies building or buying AI tools of their own.
Anthropic has provided very few specific details about what it hopes to accomplish in the drug development space. At the event, Kauderer-Abrams didn’t say what the company would do if it finds any promising drug candidates. Anthropic did not respond to The Verge’s requests for comment seeking more details, including what diseases it plans to target first and whether it would partner up with other companies for lab work, animal testing, clinical trials, or manufacturing.
AI is applied at “every single stage of drug discovery.”
Experts told The Verge that the uncertainty surrounding Anthropic’s plans reflects a broader uncertainty around the AI drug boom itself. “AI drug discovery” can mean many things. It “is a really broad term,” explained Namshik Han, a professor at the University of Cambridge and cofounder of AI biotech startup CardiaTec. AI is applied at “every single stage of drug discovery,” he said, from finding new compounds and improving them to supporting research, data analysis, clinical trials, and even manufacturing. Every major drug company will be using AI in some way, he said. Matthew Todd, a professor of drug discovery at University College London, echoed the sentiment that AI already pervades drug discovery and research, calling it a “catchall phrase” given its broad array of uses.
AI is undoubtedly changing drug development. Han pointed to the numerous initiatives by pharma giants like AstraZeneca, Novo Nordisk, and GSK, and said AI can already help generate possible drug ideas, such as by suggesting new molecules that could interact with parts of the body like cell receptors that are already known to be involved with a particular disease or are targets of existing drugs. Todd said it’s immensely useful for speeding up research and helping “road test” new drug ideas. Given Anthropic’s work on frontier models, the company would presumably use generative AI to search across vast chemical and biological possibilities and help researchers make connections that would be difficult or slow to find otherwise, potentially suggesting new drug ideas, identifying new disease targets, or finding new uses for existing drugs.
But that is still a long way from an AI-designed drug reaching patients. Todd said the field is “a long way off” from an AI-designed drug being approved by regulators for human use. He added that the drug discovery process would not run autonomously, with human input and supervision required throughout. Todd and Han both noted the lack of publicly available, high-quality experimental data, such as how various chemicals behave in the body, could slow drug development efforts as well, stressing that even for well-studied areas of biology there are still large gaps in our understanding of how things work.
AI models “haven’t yet come close to making experiments unnecessary.”
AI is not positioned to fix many of the slowest parts of drug discovery. Frank von Delft, a professor of structural chemical biology at the University of Oxford and head of protein crystallography at the Oxford Centre for Medicines Discovery, said people are right to get excited about advancing AI models, but they “haven’t yet come close to making experiments unnecessary.” Drug candidates still have to be tested in the real world for efficacy, toxicity, and whether they have practical properties allowing them to be prepared, stored, and delivered safely as medicines. All of that requires skilled workers, a lot of money, and time, especially clinical work in humans — a point when many promising drug candidates fail. If Anthropic wants to develop a drug, von Delft said, it is “going to have to spend a lot on experiments.”
It’s possible Anthropic is willing to try. In the last year, the company has been actively hiring biologists and building its own wet labs, and as of writing it has several live applications hiring for life sciences roles. Han said Anthropic has been “actively recruiting” in the area too, adding that several of his academic colleagues had been approached by the company. Without naming names, Han said he thinks Anthropic has successfully hired a few candidates away from Big Pharma and prestigious academic institutions.
With all of this complexity, whatever disease Anthropic picks, any payoff is likely a long way away — at the very least, the better part of a decade, given how long it typically takes a new drug to go through clinical trials. There’s “always a big lag time” with testing medicine, Todd said. “It takes time to show experimentally that something’s safe.” No AI-designed drug has yet made it through clinical trials and FDA approval to reach market. Some AI-developed candidates have entered clinical trials, but it’s hard to know how much AI contributed, where in the process it was used, or whether those candidates outperform conventional drugs. AI can speed up part of the search, but drugs still need to prove themselves the old-fashioned way: in slow, methodical experiments that take place in the real world.