Artificial intelligence company Anthropic recently announced that its AI model Claude successfully identified a previously uncharacterized new enzyme system in massive DNA sequence data, with a repetitive sequence arrangement remarkably similar to the well-known CRISPR gene editing tool. This breakthrough not only marks the first achievement from Anthropic's newly established bio-laboratory, but also caused strong reactions in the capital market, with several gene editing-related stocks declining on the same day.
In the specific discovery process, researchers assigned Claude a complex task of searching for retrotransposon systems in massive DNA databases. During a 21-hour deep search, approximately 950 Claude agents consumed about 210 million tokens. During this time, one agent detected a unique and regular sequence of repeated DNA adjacent to a certain retrotransposon gene. After removing interference, comparing known systems, and retrieving relevant literature, the AI finally submitted a detailed analysis report to human researchers.
After further analysis and experimental validation by human scientists, the research team officially named this new combination "Array-Associated Retrotransposon System" (ART), which mainly exists in bacteriophages. Structurally, the system includes a retrotransposon gene, adjacent accessory genes, and a set of non-coding repetitive DNA sequences. Although the retrotransposon itself is not unfamiliar, the complete system composed of it along with specific non-coding DNA sequences and unknown function accessory proteins had not been fully recognized in the scientific community before.
However, this AI-driven discovery is still in its very early stages. Anthropic clearly stated that researchers have not yet determined the core main function of the system, nor have they proven that it has the programmable ability to make targeted modifications to DNA like a mature CRISPR tool. Additionally, during the review, the research team found that when the same search task was run 10 times, although almost every run touched the ART-related gene locus, due to the enormous search space and the uncertainty of the agent's behavior, subsequent runs missed the upstream repetitive DNA sequences. This indicates that AI demonstrates strong capabilities in mining scientific clues, but there is still room for improvement in the stability of the discovery process.
Although its practical application value remains to be verified, this discovery has already attracted widespread attention in the field of gene editing. The CRISPR gene editing tool originally began as a special repetitive sequence in bacterial DNA, and now has evolved into the cornerstone of modern biotechnology. Several industry pioneers, including Professor Feng Zhang from the Massachusetts Institute of Technology and the Broad Institute, have given high praise to this achievement, considering it an extremely inspiring example of AI assisting in cutting-edge biological discoveries. As Anthropic continues to advance the exploration of using AI to propose hypotheses and assist humans in wet experiments, the empowerment boundaries of artificial intelligence in the life sciences will be further expanded in the future.