Large Genome Models Used to Design Functional New Viruses
Researchers have successfully utilized large genome models, such as Stanford's Evo, to design functional and genetically distinct versions of bacteriophages. This marks a significant milestone where generative AI models created viable, synthetic viral genomes from scratch. This breakthrough demonstrates the potential of generative AI in synthetic biology, offering new pathways for phage therapy to combat antibiotic-resistant bacteria. It also highlights the rapid advancement of genomic engineering, while raising important biosecurity considerations regarding the creation of synthetic viruses. In testing, the generative AI models designed novel genetic architectures that differed significantly from natural templates like the phiX174 virus. Although only a fraction of the generated designs (specifically 16 out of 302) proved viable in laboratory testing, it proves that AI can generate functional, complex biological systems.
## BACKGROUND
Bacteriophages are viruses that infect and replicate within bacteria, making them highly valuable tools for treating bacterial infections, especially as antibiotic resistance rises. Large genomic models function similarly to large language models (LLMs), but instead of processing human languages, they are trained on the "language" of DNA, treating nucleotide bases (adenine, cytosine, thymine, and guanine) as words to predict and generate genetic sequences.