AI Breakthrough in De Novo Viral Design
In a landmark achievement, a team of researchers from Stanford University and the nonprofit Arc Institute has leveraged artificial intelligence to design and create entirely new viruses from scratch. This represents the first instance where AI has generated complete, functional viral genomes that are effective in real-world biological tests. The work, led by computational biologist Brian Hie of Stanford, utilized a genomic large language model named Evo to propose novel genetic codes for viruses.
The researchers fine-tuned existing genomic language models, specifically Evo 1 and Evo 2, on a vast dataset of viral genomes, including 14,500 viruses from the Microviridae family of bacteriophages. These models, which are related to transformer architectures but trained on DNA sequences rather than text, learned to generate chains of nucleotides, the fundamental building blocks of DNA. By prompting the models with the initial part of a known virus's genome, they were able to generate entire novel genomes.