Early progress on accelerating drug development with Claude. Anthropic's Life Sciences team is working to accelerate the development of ...
 |
| Early progress on accelerating drug development with Claude. |
AI's Role in Drug Development
Artificial intelligence offers numerous opportunities to accelerate drug development, extending far beyond its commonly discussed applications, such as proposing initial molecular designs. Drug development involves a wide range of complex activities. These include conducting preliminary research to identify suitable therapeutic targets, determining the appropriate type of drug to interact with those targets, designing molecules with specific properties, optimizing manufacturing processes for reliable large-scale production, preparing regulatory documentation for clinical trials, and designing and administering clinical studies.
Anthropic's strategy for advancing AI-driven medical innovation involves integrating artificial intelligence increasingly into each of these activities. Rather than focusing on a single problem, the company recognizes that obstacles and inefficiencies exist throughout the entire drug development process. Identifying and addressing these challenges requires direct involvement in real-world drug development programs across different disease areas and stages of research.
Consequently, Anthropic has established partnerships with six organizations and initiated independent drug discovery programs, including research into rare childhood forms of muscular dystrophy, where Claude performs a substantial portion of the scientific and operational work. These initiatives serve three important objectives. First, they enable Anthropic to evaluate the strengths and limitations of its AI models across a broader range of drug development activities.
The findings have already informed specific priorities for improving the company's models. Second, the projects provide practical case studies for the wider life sciences community. Demonstrating how Claude can contribute to developing new treatments may encourage more biotechnology companies to integrate AI into their research and development processes. Third, the programs have the potential to produce direct medical benefits. If their progress continues positively, some projects could result in effective treatments for conditions including food allergies, rare childhood diseases, and several forms of cancer.
Case Studies
Generate Biomedicines is investigating ways to expand the therapeutic applications of CAR T-cell therapy, a cancer treatment historically used primarily against blood cancers. The company's objective is to reengineer this therapy to make it effective against solid tumors. Its approach involves shortening one section of the therapy's genetic instructions, creating additional space for instructions designed to target other types of cancer. Six Generate Biomedicines scientists used Claude Code to propose alternative designs for this genetic component.
One proposed design was approximately 70% shorter than the standard version and demonstrated improved performance in laboratory testing. It enabled immune cells to produce approximately 10 times as many receptors, the proteins responsible for recognizing cancer cells. The modified cells also demonstrated more efficient tumor-cell destruction and remained active for longer periods. These early findings demonstrate how AI-assisted genetic engineering could contribute to improving existing cancer therapies.
Divergence Labs is developing an innovative treatment approach for food allergies. Food allergies are triggered by immune reactions involving immunoglobulin E (IgE). Some IgE molecules are attached to immune cells, while others circulate freely in the bloodstream. Unlike conventional medications that primarily neutralize circulating IgE, Divergence Labs is developing drugs designed to detach IgE molecules completely from immune cells. This approach could potentially provide more effective treatment. The proposed therapeutic design requires two precisely engineered components: one that binds strongly to IgE and another that helps remove IgE from immune cells.
Claude translated Divergence Labs' research objectives into a detailed design strategy and subsequently generated approximately 35,000 antibody designs, largely autonomously, within about one month. During preliminary testing, researchers identified promising candidates for both components. Three designs demonstrated the ability to bind to IgE, while 20 showed the ability to dislodge it from immune cells. Among the latter group, three candidates appeared to perform better than the standard medication. The next development stage involves combining the two components into a single therapeutic drug.
BigHat Biosciences combines artificial intelligence with automated laboratory systems to improve potential drug candidates, commonly referred to as leads. These candidates must demonstrate several essential characteristics, including strong target binding, stability throughout manufacturing and storage, and resistance to unwanted molecular aggregation. Scientists from BigHat Biosciences and Anthropic worked with Claude to improve three properties in two antibody-fragment candidates.
One candidate was designed to target cancer cells, while the other was intended to extend the duration a drug remains in the body. Connected to BigHat's laboratory data system, Reccy, Claude proposed molecular modifications that were subsequently synthesized and tested by the company's laboratory. Approximately 1,010 molecular variants were produced and evaluated over approximately 10 weeks. During the research, the team identified potential limitations in conventional heat-stability and aggregation measurements.
Claude and an Anthropic researcher therefore reanalyzed the underlying experimental data and proposed alternative evaluation metrics and a modified testing protocol. BigHat's scientists refined and implemented the proposed approach, and the primary predictions were supported by subsequent testing. Compared with the original candidates, the best-performing versions demonstrated more than 10 times stronger binding affinity. The cancer-targeting candidate remained in solution at temperatures 18°C higher, while the candidate designed for longer-lasting drugs aggregated more than 100 times more slowly. These findings illustrate how AI can contribute to molecular optimization, experimental analysis, and the refinement of laboratory methodologies.
Enveda faced a documentation challenge. Drug manufacturing and clinical development require extensive documentation demonstrating that every production batch meets strict quality standards. Pharmaceutical companies must also prepare regulatory applications and supporting records before conducting clinical trials. At Enveda, manufacturing chemists previously spent up to 75% of their working time completing paperwork. To address this administrative burden, Enveda developed a system connecting its manufacturing documentation to Claude, enabling the AI model to identify potential inconsistencies and compliance issues.
During evaluations involving 1,061 pages of documentation, Claude identified all 11 issues previously detected by human reviewers and uncovered an additional seven that had been overlooked. The system reduced human review time by 89%, decreasing the process from more than 25 hours to less than three hours. Claude also prepared initial drafts of two important regulatory documents. One document was completed in approximately four hours, compared with an estimated three months using conventional manual preparation. For the second document, Enveda's standard human review process identified only one minor error. These results demonstrate the potential of AI to improve pharmaceutical documentation, regulatory preparation, and manufacturing quality assurance while reducing administrative workloads.
Drug discovery is often slow, expensive, and associated with significant uncertainty. Consequently, biotechnology investors have traditionally favored established research approaches with previously validated outcomes over more ambitious experimental programs, even when innovative approaches could potentially benefit more patients. Anthropic has collaborated with The Column Group (TCG) to investigate whether Claude can reduce the time and financial resources required for drug discovery, enabling investors to support more ambitious medical research initiatives.
As part of this collaboration, TCG is establishing Pleiades, a biotechnology company designed to use Claude to manage a substantially larger number of drug development programs, including more complex research initiatives than would typically be feasible for a small team. The initiative explores how artificial intelligence could reduce the scientific, operational, and administrative resources required to establish and operate biotechnology companies. If successful, this model could be replicated across the biotechnology industry, potentially directing more investment toward challenging medical conditions that have historically received limited research attention. Pleiades is currently developing specialized tools and workflows for Claude. Its first Claude skill, designed to evaluate whether particular drug targets warrant further investigation, has been assessed using 45 target-disease pairs.
Anthropic has also begun collaborating with Bristol Myers Squibb (BMS) to identify and address inefficiencies in clinical trial planning. BMS is using Claude to evaluate clinical development plans for an early-stage drug and estimate the probability of successful clinical trials. The collaboration is expected to expand in 2027, with the objective of examining how AI can improve the likelihood of success during early-stage drug development.
According to Michael Pourdehnad, MD, Senior Vice President and Head of Early Clinical Development, Hematology, Oncology and Cell Therapy at Bristol Myers Squibb, Claude has been used to analyze proprietary data, develop clinical study plans, and evaluate the likelihood of study success. He explained that Claude generated a comprehensive preliminary analysis from a single prompt, helping inform the assessment of different development options that would otherwise require several days of work. The collaboration also demonstrated that sensitive proprietary information could be handled while maintaining the required data security protections. BMS intends to expand the application of Claude across its development pipeline to support broader portfolio-level decisions.
Drug Discovery Research
Beyond its external partnerships, Anthropic has initiated independent drug discovery programs. The company acquired Mithril Medicine, a one-person biotechnology company that had been developing treatments for two rare forms of muscular dystrophy. Mithril Medicine's founder, Cody Tranbarger, had used Claude to coordinate contracts, manage logistics, organize experiments involving 11 external laboratories, and delegate numerous operational responsibilities. Through this approach, Mithril advanced a potential treatment for one form of muscular dystrophy from an initial concept to promising results in mice in less than 18 months. Following the acquisition, Cody Tranbarger joined Anthropic and continued the research, with plans to initiate several additional drug development programs. If the drug candidates continue to demonstrate promising results, Anthropic expects that a potential treatment could be ready to enter human clinical trials as early as late 2028.
AI's Role in Life Sciences
Although these developments demonstrate promising progress, the research remains at an early stage. Most of the programs are still preclinical, meaning that their candidate drugs have generally not yet reached the stage of testing in humans. Anthropic intends to expand Claude's involvement in clinical trials as the existing programs advance and as additional partnerships are established with organizations already conducting clinical-stage research. The company's broader objective is to make its AI models accessible to as many drug development researchers as possible.
Several initiatives support this objective. Anthropic has introduced Claude Science, an AI-powered research workbench designed to support scientific investigation. The company is also providing Claude credits to life scientists to facilitate the use of AI in research. Additionally, Anthropic has established the Life Sciences Verification Program, which enables trusted organizations to access its most capable models under more permissive versions of its safeguards. These safeguards are designed to prevent misuse but can sometimes restrict legitimate biological research activities. The company has also launched an open call for guest investigators in the life sciences, inviting research groups to submit proposals for using Claude to investigate important questions in drug discovery and other scientific fields.
Through these initiatives, Anthropic aims to strengthen collaboration with biotechnology organizations, expand AI-assisted scientific research, and accelerate the development of new medical treatments. The company's ongoing work demonstrates how artificial intelligence can contribute across the entire pharmaceutical development lifecycle, from identifying therapeutic targets and designing molecules to optimizing laboratory experiments, preparing regulatory documentation, and supporting clinical trial decisions.
