How Scientists Are Using Claude to Accelerate Research and Discovery

Anthropic reports that scientists are increasingly using Claude as a research partner, moving beyond simple tasks to automate complex workflows, interpret massive datasets, and generate novel hypotheses across life sciences. These advanced applications are accelerating scientific discovery by eliminating bottlenecks, handling tasks that require deep knowledge, and enabling entirely new research approaches.

The Evolution of AI in Scientific Research

Claude is evolving into a comprehensive scientific collaborator, capable of working across all stages of the research process, from experimental design to data analysis. The use of Claude is reshaping scientific workflows by making them easier and more cost-effective, often compressing projects that traditionally took months into hours.

Anthropic has invested heavily in making Claude capable for scientific work, with Opus 4.5 showing significant improvements in figure interpretation, computational biology, and protein understanding benchmarks. This progress is informed by partnerships with researchers in academia and industry, supported by the AI for Science program which provides free API credits to leading researchers.

Biomni: A General-Purpose Biomedical Agent

Biomni is an agentic AI platform that solves the fragmentation bottleneck in biological research by consolidating hundreds of tools, software packages, and datasets into a single, navigable system. Researchers interact with Biomni using plain English requests, and the Claude-powered agent automatically selects the appropriate resources to execute the task.

Accelerating Genome-Wide Association Studies (GWAS)

In a genome-wide association study (GWAS), the primary challenge is not the genomic scanning but the subsequent analysis, which involves cleaning messy data, controlling for confounding variables, and interpreting genetic 'hits.' A single GWAS traditionally takes months due to the manual, multi-tool nature of the process.

In an early trial of Biomni, a GWAS that typically takes months was completed in 20 minutes. The system has been validated through several case studies:

Molecular Cloning: Biomni designed a molecular cloning experiment whose protocol and design matched that of a postdoc with over five years of experience in a blind evaluation.
Wearable Data Analysis: Biomni analyzed data from over 450 wearable files (including continuous glucose monitoring, temperature, and physical activity) from 30 different people in 35 minutes, a task estimated to take a human expert three weeks.
Gene Activity Analysis: Biomni analyzed gene activity data from over 336,000 individual cells from human embryonic tissue, confirming known regulatory relationships and identifying new transcription factors previously unconnected to human embryonic development.

Biomni includes guardrails to detect when Claude deviates from the intended path. When the default Claude approach differed from a clinician's diagnostic process, researchers taught the agent the expert methodology step-by-step, effectively encoding tacit knowledge as a system skill.

MozzareLLM: Automating Gene Knockout Interpretation

MozzareLLM is a Claude-powered system designed to automate the interpretation of large-scale gene knockout experiments, which traditionally requires hundreds of hours of manual literature review by a human expert. The system takes clusters of genes and performs the same interpretive tasks as a seasoned scientist.

Streamlining CRISPR Data Analysis

When scientists use CRISPR to knock out thousands of genes across millions of human cells, they generate massive amounts of image data. Software can group genes that produce similar damage patterns, but interpreting why these genes cluster together requires deep human knowledge of biological processes.

MozzareLLM accelerates this process by performing the following functions:

Identifying the biological processes that a cluster of genes might share.
Flagging whether genes in the cluster are well-understood or poorly studied.
Highlighting specific genes that warrant further investigation.

Cheeseman, the lead researcher, noted that Claude consistently catches findings he had missed, stating, “Every time I go through I’m like, I didn’t notice that one! And in each case, these are discoveries that we can understand and verify.”

Crucially, MozzareLLM provides confidence levels in its findings, allowing experts to make informed decisions about where to invest further research resources. Furthermore, Claude outperformed other tested AI models in one instance by correctly identifying an RNA modification pathway that other models dismissed as random noise.

Lundberg Lab: AI-Led Hypothesis Generation

While some labs focus on automating interpretation (like MozzareLLM), the Lundberg Lab focuses on solving the earlier bottleneck: deciding which genes to target in focused, high-cost screening experiments. Their system uses Claude to flip the traditional approach from educated guessing to property-based prediction.

Predicting Target Genes Based on Molecular Properties

Instead of relying on human intuition and literature reviews to select candidate genes, the Lundberg Lab built a comprehensive map of every known molecule in the cell (proteins, RNA, DNA) and their relationships. This map includes which proteins bind together and which molecules are structurally similar.

When provided a target—such as a specific cellular structure or process—Claude navigates this molecular relationship map to identify candidate genes based on their inherent biological properties and relationships.

Currently, the lab is testing this approach on primary cilia, a topic with limited existing research. The test involves comparing the gene candidates generated by Claude against those generated by human experts using the conventional spreadsheet method. If Claude can identify a significantly higher percentage of correct genes, the approach proves superior. If successful, this method could become a standard first step in focused perturbation screening, allowing labs to make informed, data-driven bets instead of relying on intuition or brute-force screening.

Conclusion: The Future of AI in Discovery

These diverse systems demonstrate that AI is rapidly becoming a research partner capable of far more than basic tasks like summarization or code generation. The usefulness of these tools grows in concert with AI capabilities; where earlier models were limited to simple tasks, more powerful agents are beginning to replicate the complex analytical work described in scientific papers.

As AI models continue to grow more intelligent, the process of scientific discovery is developing alongside them, leading to increased efficiency and the potential for novel insights.