Researchers at Stanford University have successfully designed complete bacteriophage genomes using generative AI, then synthesized and tested them in the laboratory, demonstrating how artificial intelligence can move from computational design to experimentally validated biological systems. The breakthrough, published in Science, shows that AI-generated phages can not only function in living cells but also overcome bacterial resistance mechanisms, potentially opening new avenues for treating difficult-to-cure infections.

How Did Researchers Move AI Designs Into the Wet Lab?

The Stanford team used Evo 2, a generative AI model developed in part at Stanford, to produce novel phage genomes based on ΦX174, a bacteriophage that naturally infects Escherichia coli. The process generated thousands of possible designs, but synthesizing and testing every candidate would be prohibitively expensive. To solve this problem, first author Samuel King developed a computational framework that evaluated the generated genomes against specific design criteria and narrowed the candidates before chemical synthesis.

The researchers then transferred the selected genomes into bacteria and experimentally tested them. Nearly 300 generated phages were synthesized and tested, with 16 performing particularly well against E. coli. This workflow illustrates how AI can be integrated into research to prioritize experiments rather than replace hands-on validation, a critical distinction for lab managers overseeing similar projects.

"With the cost of DNA synthesis still quite high, the framework allowed the team to concentrate on the most viable candidates," explained Brian Hie, senior author of the study.

Brian Hie, Senior Author, Stanford University

Why Does Overcoming Phage Resistance Matter for Medicine?

One of the study's most promising findings involved combining the 16 generated phages into a cocktail. When tested against E. coli strains that were already resistant to the native ΦX174 phage, the diverse cocktail rapidly overcame that resistance. This is significant because it demonstrates a potential strategy for preventing bacteria from developing immunity to phage-based treatments.

It is important to note that the experiment demonstrated the ability to overcome phage resistance, not antibiotic resistance in the tested E. coli. However, the work could contribute to longer-term efforts to develop phage-based approaches for difficult-to-treat bacterial infections, particularly as antibiotic resistance becomes an increasingly urgent global health challenge.

Steps to Integrate AI-Driven Genome Design Into Research Workflows

• Establish Design Criteria: Define clear computational benchmarks and biological requirements before generating candidates, allowing AI models to optimize designs against measurable targets rather than producing unlimited variations.
• Implement Candidate Prioritization: Develop algorithms that evaluate generated genomes against cost, feasibility, and predicted performance, reducing synthesis and testing expenses while maintaining experimental rigor.
• Maintain Experimental Validation: Ensure wet-lab teams determine whether AI-designed sequences function as intended, preserving the critical handoff between computational prediction and reproducible biological results.
• Plan for Operational Scaling: As AI-driven workflows expand, allocate resources for instrument time, staffing, and throughput management, since candidate prioritization directly affects synthesis spending and lab capacity.

The Stanford researchers have made Evo 2 openly available to the research community, though they acknowledge that increasingly capable genome-design tools introduce safety and biosecurity considerations. The team is now investigating more genetically novel phages and longer, more complex DNA sequences, suggesting that AI-driven biological design will continue to advance rapidly.

For research lab leaders, this advance points toward a future in which computational models generate biological candidates, algorithms help prioritize them based on feasibility and cost, and wet-lab teams determine whether those designs function as intended. As AI-driven laboratory workflows expand, managing the handoff between computational design and reproducible experimental validation will become an increasingly important operational consideration for institutions seeking to leverage these tools effectively.

The implications extend beyond phage therapy. The workflow demonstrates a broader principle: AI can accelerate the early stages of biological design by generating thousands of candidates and filtering them intelligently, but human expertise and experimental validation remain essential. This hybrid approach may become a template for other areas of synthetic biology, from enzyme engineering to vaccine development, where AI can propose novel sequences but laboratory testing determines which designs actually work in living systems.