Generative design of novel bacteriophages with genome language models
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (PREreview)
Abstract
Many important biological functions arise not from single genes, but from complex interactions encoded by entire genomes. Genome language models have emerged as a promising strategy for designing biological systems, but their ability to generate functional sequences at the scale of whole genomes has remained untested. Here, we report the first generative design of viable bacteriophage genomes. We leveraged frontier genome language models, Evo 1 and Evo 2, to generate whole-genome sequences with realistic genetic architectures and desirable host tropism, using the lytic phage ΦX174 as our design template. Experimental testing of AI-generated genomes yielded 16 viable phages with substantial evolutionary novelty. Cryo-electron microscopy revealed that one of the generated phages utilizes an evolutionarily distant DNA packaging protein within its capsid. Multiple phages demonstrate higher fitness than ΦX174 in growth competitions and in their lysis kinetics. A cocktail of the generated phages rapidly overcomes ΦX174-resistance in three E. coli strains, demonstrating the potential utility of our approach for designing phage therapies against rapidly evolving bacterial pathogens. This work provides a blueprint for the design of diverse synthetic bacteriophages and, more broadly, lays a foundation for the generative design of useful living systems at the genome scale.
Article activity feed
-
You can read the final King et al 2026 paper here: www.science.org/doi/10.1126/... It was first pre-printed in Sep 2025: www.biorxiv.org/content/10.1...
-
Here is the preprint (2025) Generative design of novel bacteriophages with genome language models Samuel H. King, Claudia L. Driscoll, David B. Li, Daniel Guo, Aditi T. Merchant, Garyk Brixi, Max E. Wilkinson, Brian L. Hie bioRxiv 2025.09.12.675911
-
Preprint’s been out for nearly a year, BTW www.biorxiv.org/content/10.1...
-
here is the state of the art. the phage design pipeline shown here does not even produce sequences with enough novelty to dodge the similarity screening done by DNA synthesis houses. then you have to actually build and test your designs in the lab
-
www.biorxiv.org/content/10.1...
-
"Generative design of novel bacteriophages with genome language models" #BioInformatics #GenAI #Genome #LanguageModel #ReSearch ... En bref : génération et tests de génomes générés à partir de ceux connus : nouvelles découvertes ! www.biorxiv.org/content/10.1...
-
Some Stanford biotech guys did it as a stunt, yeah. www.biorxiv.org/content/10.1...
-
The paper, “Generative design of novel bacteriophages with genome language models”
-
Do not know nearly enough about AI but thought this phage design preprint was cool. www.biorxiv.org/content/10.1...
-
Paper for today’s discussion is “Generative design of novel bacteriophages with genome language models” www.biorxiv.org/content/10.1...
-
Those models are called GLMs (genome language models) and they're quite good actually, they just have (as usual) very little to do with the LLM hype these CEOs are frothing at the mouth over. www.biorxiv.org/content/10.1...
-
www.biorxiv.org/content/10.1...
-
You can read the paper here:
-
The paper of record is too lazy to find and read a preprint to see they just did what academics with LLMs have done for over a decade now. www.biorxiv.org/content/10.1...
-
Generative design of novel bacteriophages with genome language models https://lobste.rs/s/yvqwqc #science #vibecoding
-
Don't ask Claude Sonnet 4.5 about this paper: www.biorxiv.org/content/10.1... It's about using genome language models to create new bacteriophages and I swear to god it triggers an ASL-3 chat lockdown.
-
🤖 New groundbreaking research from @brianhie.bsky.social, Samuel King, and team, leverages genomic foundation models like Evo2 to generate viable bacteriophage genomes, pushing the boundaries of synthesis at scale combined with GenAI, powered by Twist DNA: www.biorxiv.org/content/10.1...
-
Preprint here: www.biorxiv.org/content/10.1...
-
Who had “AI Super Virus” on their apocalypse bingo card? #Blackout #Bingo www.biorxiv.org/content/10.1...
-
what is he talking about look how many new viruses we have already! www.biorxiv.org/content/10.1...
-
www.biorxiv.org/content/10.1...
-
➡️ Die Forschungsarbeit von Forschenden des Arc Institute und der Stanford University, Palo Alto, 🇺🇸, auf dem Preprint-Server "bioRxiv" www.biorxiv.org/content/10.1... 2/2
-
🫴 AI-generated virus: Generative design of novel bacteriophages with genome language models: www.biorxiv.org/content/10.1... #ΦX174
-
I dunno, using an LLM to design new viruses and then *actually generating the viruses* seems like it might be a bad idea? www.biorxiv.org/content/10.1...
-
This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/17178008.
This manuscript describes very cool molecular biology with careful engineering and a convincing structural and functional assessment of a reconstructed ΦX174-like phage. The build-test workflow is clearly described, and the recovery of infectious particles provides a strong experimental anchor for the computational design story. This is a great demonstration of what is currently feasible for whole-genome phage engineering.
However, it misses a major opportunity to contextualize the generative model performance with explicit baselines. This makes it difficult to assess whether there is a true advance due to the ability of the generative model to move in sequence space with success rates …
This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/17178008.
This manuscript describes very cool molecular biology with careful engineering and a convincing structural and functional assessment of a reconstructed ΦX174-like phage. The build-test workflow is clearly described, and the recovery of infectious particles provides a strong experimental anchor for the computational design story. This is a great demonstration of what is currently feasible for whole-genome phage engineering.
However, it misses a major opportunity to contextualize the generative model performance with explicit baselines. This makes it difficult to assess whether there is a true advance due to the ability of the generative model to move in sequence space with success rates (~16/300) above alternative (lightweight) models. Moreover, the sequence space explored here appears to stay close to known natural sequences, so it is hard to appreciate how the generative model adds novelty beyond proximity to existing genomes. Side-by-side comparisons against simple alternatives (for example a consensus design, frame-preserving randomization under GC and codon constraints, ancestral reconstruction, or family statistical models such as direct coupling analysis) would clarify whether the model improves assembly success or shifts resistance outcomes. Recent studies provide useful templates for such "null model" panels, see OpenCRISPR Figure 3E (https://pmc.ncbi.nlm.nih.gov/articles/PMC12422970/), and ProGen Lysozymes Figure S10 (https://pmc.ncbi.nlm.nih.gov/articles/PMC10400306/) for evaluating LLM-generation of proteins. Including analogous baselines here, with success rates and resistance readouts reported for each, would make the impact of this contribution and generality of this approach unmistakable.
Competing interests
J.S.F. receives consulting fees from and has an equity interest in Profluent Bio, a company using generative AI. J.B.-D. is a scientific advisory board member of SNIPR Biome, Excision Biotherapeutics, LeapFrog Bio, and Acrigen Biosciences and is a co-founder of Acrigen Biosciences and ePhective Therapeutics.
-