Anthropic partners with biotech firms to accelerate drug discovery

A multi-stage AI strategy

Anthropic is moving beyond simple tasks like proposing initial designs for new molecules. Their life sciences team is working to integrate the Claude model into the entire lifecycle of drug development. This includes identifying targets, designing molecules, optimizing them for manufacturing, and drafting regulatory filings.

The company believes that the bottlenecks in medicine are distributed across many different tasks rather than one single problem. To address this, they have engaged in deep research partnerships with six different organizations. These collaborations allow them to see how the model performs in real-world programs across various disease areas.

The goal is to scale this work substantially to increase the rate of new cures. Anthropic is already running its own drug discovery programs, such as those for rare childhood forms of muscular dystrophy. These projects serve as case studies to help the wider life sciences community understand how to use AI effectively.

Engineering better therapies

One partnership with Generate Biomedicines focuses on improving CAR T-cell therapy for cancer. This therapy currently works well for blood cell cancers, but the team wants to reengineer it to treat solid tumors. They needed to shorten the genetic instructions of the therapy to make room for new instructions.

Six scientists used Claude Code to propose new designs for this specific part of the therapy. One design was about 70% shorter than the standard version and proved more effective in lab tests. It enabled immune cells to produce about 10 times more receptors while killing tumor cells more efficiently and remaining active longer.

Another project with Divergence Labs aims to treat food allergies by detaching IgE molecules from immune cells. This requires a precise two-armed design that can both bind to and dislodge the IgE. Claude converted the goals into a design plan and then autonomously generated roughly 35,000 antibody designs in about a month.

Optimizing lab results

BigHat Biosciences uses AI paired with an automated lab to improve potential drug leads. These leads must bind tightly to targets, remain stable during storage, and avoid clumping together. Scientists worked with Claude to improve these three properties for both cancer-targeting and long-lasting drug fragments.

Claude was connected to a lab data system called Reccy to propose changes to each molecule. The lab then manufactured and tested about 1,010 variants in about 10 weeks. When the team suspected the original metrics for stability were misleading, Claude helped reanalyze the raw data and proposed a new testing protocol.

The results of these tests were significant for the drug leads. The best versions bound more than 10x more tightly than the originals. Additionally, the cancer-targeting lead remained stable at a temperature 18°C higher, and the other lead clumped more than 100x more slowly.

Automating the paperwork

A major hurdle in drug development is the administrative burden of manufacturing and regulatory filings. At Enveda, manufacturing chemists previously spent up to 75% of their time on paperwork. To solve this, Enveda built a system that connects manufacturing documents to Claude to flag potential issues.

In tests on 1,061 pages of documentation, Claude identified all 11 issues that human reviewers had spotted. It also flagged seven issues that the humans had missed. This method reduced the human review time by 89%, cutting the process down from over 25 hours to less than three.

Beyond auditing, Claude also wrote the first drafts of two important regulatory documents. One of these drafts took only four hours to complete, a task that would have taken about three months to do by hand. This demonstrates how the model can handle the high-volume, high-accuracy writing required for clinical trials.

The path to clinical trials

Anthropic is also exploring how AI can change the business model of biotechnology. They are working with The Column Group to establish a new company called Pleiades. This company will use Claude to run more complex drug programs than a small team could normally handle, potentially lowering the cost for investors.

In another partnership, Bristol Myers Squibb is using Claude to analyze proprietary data and estimate the likelihood of success for early-stage drugs. Michael Pourdehnad, MD, noted that Claude generated a robust initial analysis from a single prompt.

While much of this research is currently preclinical, Anthropic plans to expand Claude's role as these programs advance. They have launched Claude Science for researchers and a verification program for trusted organizations. They are also seeking guest investigators to help explore new questions in drug discovery.