AI vs. Bacteria: A Fight for Survival 🦠🔥

CHAPTER 1: THE EVOLUTIONARY AI – EVO 2
The research initiative began with the creation of Evo 2, a generative AI model developed by Brian Hie and Samuel King. Evo 2’s core function is to generate new DNA sequences from small starting snippets of phage genomes, effectively creating entirely new viral genomes. Hie, an assistant professor of chemical engineering, and King, a bioengineering graduate student, collaborated on this project, utilizing the Dieter Schwarz Foundation Stanford Data Science Faculty Fellow program. The model’s design prioritized a single, left-to-right pass for genome generation, without any manual additions or alterations to the initial sequence. This approach allowed Evo 2 to produce thousands of candidate genomes, streamlining the process of creating viable viral genomes.

CHAPTER 2: ΦX174 – A COMPACT TEST SYSTEM
The selection of ΦX174 as the primary phage model was strategic, driven by its relatively small genome size – fewer than 6,000 base pairs – compared to the human genome’s approximately 3 billion base pairs. This compact size facilitated easier interpretation of the generated sequences, simplifying the analysis of individual genes within the 5,400-character DNA sequence. Initial laboratory testing revealed that some of Evo 2’s suggested phages exhibited superior fitness compared to native ΦX174, indicating the model’s capacity to generate effective viral genomes.

CHAPTER 3: COMPUTATIONAL SCREENING AND DNA SYNTHESIS
Samuel King developed a computational framework to refine the selection of candidate genomes before costly DNA synthesis. This framework assessed traits derived from ΦX174 and related phages, prioritizing those with the highest potential for effectiveness. The framework involved multiple steps: generating genomes with Evo 2, evaluating them against design criteria, selecting optimal candidates, chemically synthesizing them, and finally, testing their performance in laboratory assays. This approach significantly reduced synthesis costs by focusing resources on the most viable candidates, streamlining the entire process.

CHAPTER 4: PHAGE MIXTURES – COMBATING RESISTANCE
Recognizing the potential for bacteria to develop resistance to single phage treatments, the research team selected a mixture of more than one E. coli-targeting phage. Hie explained that a phage cocktail presents a significant barrier to bacterial resistance, as bacteria would struggle to develop resistance to the entire cocktail of genetically distinct phages. This strategy was successfully demonstrated in laboratory testing, rapidly overcoming resistance in E. coli strains previously immune to native ΦX174. The team envisions applying this approach to combat resistant strains of MRSA and Pseudomonas aeruginosa, leading to more effective treatment options.

CHAPTER 5: OPEN-SOURCE ACCESS AND FUTURE DIRECTIONS
Brian Hie released Evo 2 as open-source software, granting researchers access to the model and enabling them to design their own phage genomes. While this release sparked discussions regarding safety and security, Hie argued that the risks associated with a publicly available tool are less significant than those posed by readily accessible pathogens. He highlighted the potential of AI-enabled systems to support responses to pandemics and provide defense against biological threats. King emphasized the creative potential of Evo 2, opening new avenues of research, and the team plans to extend Evo 2 to longer and more complex DNA sequences, exploring genomes of small bacterial species that could be engineered for producing chemicals, medicines, or fuels.