Researchers Use AI Model Evo to Design Functional Viral Genomes

Scientists are moving past simply reading genetic code and are beginning to write it from scratch using artificial intelligence. Researchers working at the Arc Institute trained an AI model named Evo on roughly two million bacteriophage genomes. Unlike language models trained on books or articles, Evo learned entirely from biological sequence data, functioning much like a large language model but for genomics.

For this project, the research team tasked the AI with proposing variants of phiX174, a simple bacteriophage consisting of just 11 genes and about 5,000 DNA letters. The model generated novel sequences that differed significantly from anything found in nature, yet some of these engineered entities retained the ability to infect and destroy bacterial targets in the lab.

How AI-Generated Genomes Survived Biological Constraints

Creating viable viral genomes is exceptionally difficult because natural selection punishes even minute alterations. Past studies show that changing a single amino acid in a viral protein carries roughly a 20 percent chance of inactivating the virus entirely.

AI Redesign Helps Enzymes Evolve Beyond Natural Limits

However, viability surged among sequences that retained high similarity to the original phiX174 phage. Outputs sharing 98 percent sequence similarity or higher achieved a 46 percent viability rate. Many of the successful AI designs closely mirrored the original gene set, leaving the starting stretch of DNA completely untouched.

Yet, the AI also produced surprising structural departures. Some viable viruses lost an entire viral protein and compensated with modifications elsewhere in the genome, while others introduced entirely new genes or swapped out genes with those from distantly related viruses.

Expert Reactions and the Debate Over Synthetic Life

The successful generation of functional viral particles has drawn diverse reactions from researchers across the synthetic biology community. Jef Boeke, a biologist at NYU Langone Health, called the project an impressive first step toward AI-designed life and noted that the AI’s performance was surprisingly good, producing arrangements that human scientists had not considered.

New Irish-Developed Heart Valve Model May Revolutionize Cardiac Treatment

Other specialists highlighted the broader significance of the milestone. Marc Güell, a researcher at the synthetic biology lab at Pompeu Fabra University in Spain, called the study a very significant turning point because it marks the first time humanity is beginning to design biology directly on a computer. Patrick Cai, chair of synthetic genomics at the Manchester Institute of Biotechnology, added that the development suggests genome language models are beginning to grasp evolutionary design rules.

Not every pioneer in the field views the breakthrough with unalloyed optimism. J. Craig Venter, known for his early work creating organisms with synthetic DNA, characterized the method as just a faster version of trial-and-error experiments. While Venter acknowledged the technological leap, he raised grave concerns regarding potential applications to dangerous human pathogens such as smallpox or anthrax, urging extreme caution around any viral enhancement research.

3D Organoid Models in Motor Neuron Diseases: Advancements and Applications

Pathways for Phage Therapy and Scaling Challenges Ahead

Despite safety concerns, researchers emphasize that the technology holds immense promise for medicine. Medical professionals have long utilized phage therapy to combat multidrug-resistant bacterial infections, and viruses also serve as vital delivery mechanisms in gene therapy. AI-assisted design could significantly accelerate the development of specialized viral tools.

Scaling these methods to living cells, however, presents a massive technological hurdle. As Boeke observed, the complexity involved in designing living cells scales far beyond current computational paradigms, though industry figures like Ginkgo Bioworks CEO Jason Kelly argue that pursuing AI-designed cellular architecture should remain a high priority.