Recent News 78 : AI Designs Viable Bacteriophage Genomes, Combats Antibiotic Resistance

AI Designs Viable Bacteriophage Genomes, Combats Antibiotic Resistance

AI models, Evo 1 and Evo 2, have now generated functional bacteriophage genomes, demonstrating experimental validation of whole genomes designed by AI

Cryo-electron microscopy structure of a viable bacteriophage designed by Evo. [Samuel King]

DNA sequences contain the underlying instructions of life for all living organisms, but even the simplest microbial genomes are largely complex, with millions of DNA base pairs encoding the interplay of DNA, RNA, and proteins for cellular function at multiple scales, from individual pathways to whole genomes. 

“Most biological functions are not achieved by any single gene,” Brian Hie, PhD, assistant professor of chemical engineering at Stanford University and innovation investigator at Arc Institute, told GEN when describing the motivation behind building genome foundation models for generative design. “If we want to engineer more complex functions, we’ll need to break out of the single gene space and design complete genomes.” 

Yet, recapitulating the tightly orchestrated interactions between multiple genes, regulatory elements, and recognition sequences, where even a single mutation can render an entire genome nonviable, is not an easy feat. 

In a step toward whole genome design, Hie and colleagues have now demonstrated the first end-to-end generative design of 16 complete, functional, and evolutionarily novel bacteriophage genomes, offering a new path for phage-based therapies against antibiotic-resistant infections. The work was posted as a preprint on bioRxiv and has not yet been peer reviewed. 

The new bacteriophage genomes were generated using the Evo series of foundation models. Published in Science last November, Evo 1 is trained on 2.7 million prokaryotic and phage genomes and was shown to generate experimentally validated CRISPR-Cas molecular complexes and transposable systems, representing the first examples of protein-RNA and protein-DNA codesign with a language model. In a follow-up preprint, the model also demonstrated the ability to generate de novo genes, including anti-CRISPR proteins and toxin-antitoxin systems. 

Just three months later, Arc Institute, in collaboration with Nvidia, released Evo 2, described as “the largest publicly available AI model for biology to date.” Evo 2 moved beyond single-cell genomes of bacteria and archaea to include information from humans, plants, and other more complex single-celled and multi-cellular species and was trained on more than 9.3 trillion nucleotides across the tree of life.  

While experimental validation of the Evo models has mostly stayed at the level of individual genes, Hie emphasized that the newly generated bacteriophage genomes demonstrate the first experimental validation of the whole genome design concept.  

Arc Institute has since maintained big investments in data-driven AI and recently launched the inaugural “Virtual Cell Challenge,” a public competition, sponsored by Nvidia, 10x Genomics, and Ultima Genomics, with a grand prize worth $100,000 for the machine learning model that best predicts how cells will respond to genetic perturbations. 

Historic phage 

As viruses that specifically infect bacteria, bacteriophages are powerful biotechnology tools used for targeted antibacterial therapies, developing diagnostic tools via phage display technology, and engineering bacteria to create novel products.  

Notably, the study used the historic bacteriophage ΦX174 as a design template. ΦX174 was the first complete genome sequenced by Nobel Laureate in Chemistry, Frederick Sanger, PhD, and first genome chemically synthesized by Craig Venter, PhD, known for heading the private-sector enterprise, Celera Genomics, in the Human Genome Project (HGP).  

“ΦX174 is a historic phage and we wanted to continue that story by designing it,” said Samuel King, a PhD candidate in the Hie lab and first author of the preprint, in an interview with GEN. 

Traditional phage therapy has been limited to engineering components of phages that exist in nature and requires a thorough understanding of the system that is being manipulated. In contrast, an AI-guided approach using Evo aims to achieve steerable design of phages by sampling outside of natural evolution.  

ΦX174 possesses an approximate 5.4-kilobase genome containing 11 genes, at least seven regulatory elements, and two recognition sequences. While the bacteriophage has a much more complex genetic architecture than any previously AI-generated biological system, ΦX174’s relatively small genomic length and rich history of experimental work made it a tractable model for establishing whole-genome design. 

Out of approximately 300 designs that were experimentally tested, 16 phages emerged as viable with substantial evolutionary diversity that enabled a phage cocktail that rapidly overcame bacterial resistance. In addition, multiple generated phages showed increased fitness relative to ΦX174 in growth competitions and lysis kinetics. 

“The idea was whether or not the model could reach new parts of genome design space that natural evolution has not yet accessed,” King told GEN. “Even at the scale of an entire genome, the models were able to reason the different elements in the genome that need to work together in order to create something functional.” 

King highlights that these AI-designed genomes can inform a phage-based strategy to target multi-drug resistant bacteria that impact significant percentages of agricultural crops worldwide. Future directions of the study aim to design larger phage genomes, which can provide more modularity and flexibility for design, and other genomic systems, such as operons in bacteria. 

Copyright belongs to,  article taken from : https://www.genengnews.com/topics/artificial-intelligence/ai-designs-viable-bacteriophage-genomes-combats-antibiotic-resistance/