Key facts
- Large genome models, analogous to large language models, were used to design novel viruses.
- The AI-generated viruses infect bacteria and are closely related to existing bacteriophages.
- Stanford University researchers developed and tested these AI-designed viral genomes.
- 16 out of 285 synthesized viral sequences demonstrated the ability to inhibit E. coli growth.
- The most successful AI-designed viruses shared significant sequence similarity with the original ΦX174 virus.
Researchers have successfully employed large genome models, a type of artificial intelligence trained on DNA sequences, to design novel viruses capable of infecting bacteria. These AI-generated viruses, developed by a team at Stanford University, are closely related to existing bacteriophages but possess unique characteristics that would be difficult to achieve through natural evolution. The study involved training models on extensive bacterial virus DNA and then fine-tuning them with sequences specific to the Microviridae family, which includes the ΦX174 virus. By using specific prompts and applying stringent filtering criteria to ensure biological relevance and functionality, the researchers synthesized 285 potential viral sequences. Of these, 16 demonstrated the ability to inhibit the growth of E. coli, indicating they functioned as viruses. The most effective of these AI-designed viruses exhibited high sequence similarity to the original ΦX174, suggesting that while AI can introduce novel features, a degree of resemblance to known functional viruses is crucial for viability. The development raises potential concerns about the future application of similar AI technologies for designing viruses that could target vertebrates.