An artificial intelligence system found an enzyme in a gene database that no one has ever described.
Anthropic says its model independently identified a previously unrecorded enzyme structure with features characteristic of gene-editing systems. The enzyme is called array-associated reverse transcriptase, and it mainly exists in bacteriophages. It is composed of a reverse transcriptase, a partner gene, and a uniformly spaced repetitive sequence; the repetitive portion is similar to the array that enables the editing tools to be programmable.
The process is worth noting. The model received a broad instruction to search a large gene-sequence repository for anomalous reverse transcriptases. About 950 agents worked for 21 hours, consuming roughly 210 million tokens. Ultimately, one of them discovered the repeat pattern. The scale shows that the search itself has already been engineered.
This is the first result from the company’s newly built life-science lab. By handing the search and hypothesis generation to machines, human researchers can save a large amount of screening time. The bottleneck shifts from finding leads to validating them, and lab output capacity becomes the scarce resource.
The company is preparing for an IPO, so naturally, these kinds of achievements will be woven into the narrative. More meaningful for the industry is that the process itself can be replicated. Research is shifting from individual inspiration to large-scale screening.
Machines won’t suddenly get inspiration, but they can search tirelessly.
#人工智能 #生物科技
Anthropic says its model independently identified a previously unrecorded enzyme structure with features characteristic of gene-editing systems. The enzyme is called array-associated reverse transcriptase, and it mainly exists in bacteriophages. It is composed of a reverse transcriptase, a partner gene, and a uniformly spaced repetitive sequence; the repetitive portion is similar to the array that enables the editing tools to be programmable.
The process is worth noting. The model received a broad instruction to search a large gene-sequence repository for anomalous reverse transcriptases. About 950 agents worked for 21 hours, consuming roughly 210 million tokens. Ultimately, one of them discovered the repeat pattern. The scale shows that the search itself has already been engineered.
This is the first result from the company’s newly built life-science lab. By handing the search and hypothesis generation to machines, human researchers can save a large amount of screening time. The bottleneck shifts from finding leads to validating them, and lab output capacity becomes the scarce resource.
The company is preparing for an IPO, so naturally, these kinds of achievements will be woven into the narrative. More meaningful for the industry is that the process itself can be replicated. Research is shifting from individual inspiration to large-scale screening.
Machines won’t suddenly get inspiration, but they can search tirelessly.
#人工智能 #生物科技
