Generating functional plasmid origins with OriGen

While generative artificial intelligence has shown promise for biological design, no computational system has yet created sequences proven capable of replication. Focusing on plasmids as minimal replicating systems, we develop OriGen, a language model that generates novel plasmid origins of replication while maintaining essential functional elements. We experimentally validate OriGen's ability to create active origins in Escherichia coli that diverge from existing wild types, demonstrating the model's capacity to capture the complex and often enigmatic mechanisms of biological replication. Further, we directed OriGen to produce functional origins that avoid common restriction sites, demonstrating the model's potential for automated controllable design of vectors with desirable properties.