From Text to DNA: How AI Learned to Write Viruses
In a groundbreaking experiment, researchers at Stanford equipped two artificial‑intelligence models, dubbed Evo 1 and Evo 2, with the ability to predict the next nucleotide in a DNA strand. Unlike conventional chatbots that guess the next word, these systems were trained on genetic sequences, allowing them to generate entire viral genomes from scratch. The team chose the modest bacteriophage phiX174 as a template—a virus with a 5,400‑base genome and only eleven genes—providing a simple scaffold for the AI to expand upon.
Thousands of Designs, Sixteen Real‑World Winners
The algorithms churned out thousands of novel genomes. After a rigorous filtering process, roughly three hundred candidates were synthesized and introduced to laboratory cultures of gut bacteria. Sixteen of these engineered phages proved viable: they entered bacterial cells, hijacked the host machinery, and produced progeny, mirroring the life cycle of natural viruses. Intriguingly, about half of the successful strains acquired spontaneous mutations during replication, indicating that the AI’s output was further refined by evolutionary pressure.
Uncharted Genetic Territory
None of the functional phages resembled any virus previously catalogued in nature. Some carried unique insertions, such as a protein‑packing module borrowed from a distant viral lineage, effectively creating a novel “delivery system” for their genetic payload. Microscopic imaging of these synthetic particles revealed unprecedented structural features, underscoring the creative potential of machine‑generated DNA.
Targeting Resistant Bacteria
To test therapeutic relevance, the scientists exposed antibiotic‑resistant gut bacteria—specifically strains that had evolved defenses against phiX174—to a cocktail of the AI‑designed phages. The synthetic mixture swiftly eradicated the pathogens, whereas a comparable blend of naturally sourced, related phages failed to achieve the same result. All experiments were confined to petri dishes, using a single harmless bacterial strain, and no animal models or human subjects were involved.
Safety, Ethics, and the Need for Regulation
The study, published in Science, sparked a parallel commentary from Johns Hopkins University scholars. While praising the team’s thorough safety assessments, the commentators warned that the ability to design viral genomes now exists without a comprehensive regulatory framework. They advocated for mandatory screening of custom DNA orders by commercial synthesis companies—a practice that currently relies on voluntary compliance.
Looking Ahead
These findings hint at a future where AI‑crafted bacteriophages could complement or even replace traditional antibiotics, especially as bacterial resistance continues to rise. The researchers envision scaling the approach to larger bacterial genomes, potentially tackling a broader spectrum of pathogens. However, the leap from a controlled laboratory setting to clinical application will require extensive testing, ethical deliberation, and robust biosecurity measures.
Source: https://scientias.nl/ai-ontwerpt-zestien-echte-virussen-die-bacterien-doden/
