arstechnica.com
Researchers developed Evo, a genomic language model, that can predict genomic outputs based on DNA prompts. Trained on bacterial genomes, Evo can complete gene sequences and restore missing genes in functional clusters. Importantly, Evo understands evolutionary constraints, making changes primarily in variable protein regions. To test Evo’s creativity, they prompted it with a novel bacterial toxin. By filtering out responses resembling known antitoxins, the researchers aimed to see if Evo could generate entirely new, functional proteins. This demonstrates Evo’s potential to create novel genomic sequences beyond simple replication of existing data.
