ProductsMulti-source71°

Anthropic 发布 Claude 医药合作案例:从抗体设计到一人药企

Why we picked it

Anthropic 放出了一堆 Claude 帮药企干活的真实案例,有人靠它一个人跑完临床前测试,细节和完整技术报告都在。

Anthropic 公布了与多家生物制药公司合作的案例研究,展示 Claude 在药物研发各环节的应用。其中 Divergence Labs 的案例里,Claude 在底层调用蛋白质折叠模型来设计针对食物过敏的抗体,扮演的是编排器的角色。Mithril Medicine 的创始人借助 Claude 处理运营瓶颈,一人完成了杜氏肌营养不良项目的临床前测试。所有案例的验证仍停留在体外实验阶段,与 Bristol Myers Squibb 的临床试验规划合作也处于早期。作者认为这些与现有药企合作的案例,比 Anthropic 一个月前发布的酶系统发现更有实际价值。

Original · Tanishq Abraham (论文推介)

These case studies are actually really cool and impressive!

I have many thoughts :)

When people think AI will cure disease, they think Claude/ChatGPT will pop out a magical drug and that's it.

No, that's not how it works!

There is so much more needed to make a lifesaving drug available to patients, from further optimization, clinical trials, manufacturing stuff, regulatory filings, etc.

Anthropic has shared case studies on working with existing biopharma companies to accelerate these processes, with very promising results.

Most notably they are showcasing examples working with existing companies rather than their own pipeline. This is obviously the strategy that will have nearest term impact on patient lives imo.

Frankly their release a month ago on discovering that enzyme system is significantly less interesting and valuable than all of this work shared here.

That said, I think it's worth noting what types of problems they haven't used Claude for. All the validation is done in vitro, and nothing has been done preclinically or in clinically (they are doing some work with Bristol Myers Squibb for clinical trial planning but it's very early on). This is because in vitro work has a much faster validation loop!

More work is needed to either bring in vitro biology closer to human biology or directly learn human biology!

Additionally, it's worth noting that in many cases Claude is an orchestrator over various specialized tools. For example, in the Divergence Labs' case study applying Claude to design antibodies to target food allergies, Claude actually uses protein folding models under the hood. Claude isn't doing everything by itself.

Finally, I will point out the particularly interesting acquihire they did: the one-person biopharma. By letting Claude handle much of the operational bottlenecks, the founder of Mithril Medicine was able to successfully manage his drug program for muscular dystrophy all by himself. He has successfully done preclinical testing too!

AI is enabling more lean workflows in biopharma and helping to accelerate early-stage drug development!

Btw there are so many details including full technical reports for each case study! Will have to dig even deeper :)

  • Latent Space (swyx)10-27 18:40原文
  • kimmonismus10-27 21:11原文
  • Andrew Curran10-27 22:01原文
  • rohanpaul_ai10-27 20:54原文
  • @koltregaskes10-26 07:27原文
  • Gorden Sun10-26 07:31原文
  • Claude Code: GitHub Releases10-26 21:51原文
  • IT之家10-26 23:28原文
  • Anthropic10-27 22:20原文
  • Blognone10-26 01:42原文