Scientists have designed a functioning virus from scratch using AI. What you need to know

Scientists have created a functioning virus from scratch using artificial intelligence, marking a significant milestone in the field of "generative biology." The research involved designing hundreds of new versions of a well-studied bacteriophage, a virus that infects bacteria, using AI. Out of 285 AI-designed genomes tested, 16 produced working phages capable of infecting E. coli bacteria.

This breakthrough demonstrates that AI can not only analyze genetic code but also write entirely new versions of biological molecules. However, it is essential to understand what the researchers have and have not demonstrated. Bacteriophages are highly specialized predators that recognize specific molecules on a bacterium's surface, inject their genetic material, and take control of the host cell's machinery to produce new copies of themselves.

Their genomes are often small, making them ideal for testing AI capabilities. The AI models used in this study, called Evo 1 and Evo 2, work similarly to large language models but with genetic sequences. The researchers trained these models on over 2 million phage genomes and asked them to design new versions of a small, well-studied phage called ΦX174, which has an 5,400 DNA letter genome and produces 11 proteins.

The AI was not inventing a virus from scratch but using patterns found in existing biology to generate new combinations. Out of the 285 designs tested, only 16 produced functioning phages. Some of these phages behaved similarly to ΦX174 despite having different DNA sequences. The study also explored whether the new phages could overcome bacterial resistance.

After repeated exposure to a resistant variant of E. coli, hybrid phages emerged that could infect the previously resistant bacteria. While AI has generated a diverse starting population for evolution to work with, the final phages were not directly designed by the AI. The researchers emphasize that this study is a demonstration of what might be possible with AI and phages, not a proof of phage therapy for patient treatments.

The phage used in this study is unusually simple, and many clinically relevant phages have more complex genomes. There are also significant regulatory challenges for personalized phage therapy. Nonetheless, this research highlights the potential of AI in designing new biological entities and sheds light on the usefulness of phages in understanding genetics and combating antibiotic resistance.