Anthropic Discovers 'ART', a Novel CRISPR-like Enzyme System, Using Claude

Anthropic has established a new life sciences research group and introduced a novel research approach that uses the AI model Claude to identify uncharacterized protein families in DNA datasets. As an initial result, Claude autonomously discovered ART (Array-associated Reverse Transcriptases), a new enzyme system with repetitive sequence patterns similar to CRISPR.

The Discovery Process and Characteristics of ART

Researchers provided Claude with only an initial prompt to find interesting reverse transcriptases (RT) in a massive DNA sequence database. Approximately 950 Claude agents explored the data, processing 210 million tokens over 21 hours. During this process, one agent detected repetitive patterns in non-coding DNA adjacent to specific RT genes, confirming it was a new system not previously reported.

ART is primarily found in bacteriophages and consists of the following three components:

• Reverse Transcriptase (RT): An enzyme that copies RNA into DNA
• Partner Gene: An auxiliary protein of unknown function located next to the RT
• DNA Repeat Array: Non-coding DNA sequences evenly spaced similarly to CRISPR arrays

Initial experimental results showed that ART arrays are expressed as multiple short RNAs, suggesting the potential to become a programmable tool like the CRISPR-Cas system.

AI-Based Biology Research Workflow

Anthropic's research team used Claude Science and Claude Code to automate the research process. Claude performs literature reviews, data reproduction, candidate screening, and the creation of human-readable reports. In a single investigation campaign, hundreds to thousands of candidate reports are generated, and human scientists review them to select candidates worth experimenting with. The hypotheses generated in this process are used to train Claude's 'scientific taste'.

MIT Professor Feng Zhang commented on this discovery, calling it "an interesting case of how AI agents can contribute to biological discovery," and noted that the association between RNA repeat arrays and reverse transcriptases requires further investigation. Anthropic plans to release these research findings as a preprint and expand collaborations with external scientists to demonstrate the value of AI-based hypothesis generation across the life sciences.