A research team from Stanford University and the Arc Institute has successfully designed functional viral genomes from scratch using a specialized AI model. The findings, published in the journal Science, demonstrate that these AI-generated viruses possess the biological capability to effectively kill bacteria.
The research team utilized a genomic AI model dubbed "Evo," which was initially trained on a massive dataset of approximately 9 trillion nucleotides to distill the fundamental patterns of biological evolution. The model underwent further specialized training using data from the bacteriophage Phi X-174 and roughly 15,000 related species. From the 700,000 potential genomic candidates proposed by the AI, the team synthesized promising DNA sequences. Ultimately, they confirmed that 16 of these viruses—despite having structures that do not exist in the natural world—were fully functional, exhibiting robust replication and potent bacterial-killing capabilities.
The viruses produced in this study are not mere replicas of existing organisms; they demonstrate stability and replication speeds that meet or exceed those found in nature. This represents a significant milestone in AI-driven biological engineering. However, the study also highlights critical biosecurity concerns. The research team has implemented strict safeguards, such as intentionally excluding data on human-pathogenic viruses from the training set. Moving forward, this technology holds immense promise as an innovative tool for advancements in medicine and biotechnology.