Researchers at Stanford University have developed a groundbreaking AI model, named Evo, that has led to the discovery of never-before-seen proteins by analyzing bacterial genomes. This innovative approach, which resembles large language models, has the potential to revolutionize protein discovery by exploring genetic blueprints in a novel way.
The traditional focus of AI in biology has been on predicting and designing protein structures, but this new method delves deeper into the relationship between a protein’s function and its genetic origins. Evo was trained to anticipate the next base in a sequence within bacterial genomes, rewarding accurate predictions. By tapping into the intricate connections between genes at the nucleic acid level, the model was able to uncover proteins with unprecedented characteristics, challenging our existing knowledge of protein diversity.
This breakthrough was made possible by leveraging the gene clustering phenomenon commonly found in bacterial genomes. The researchers developed a ‘genomic language model’ that was able to analyze the organization of bacterial genes with related functions, leading to the prediction of novel proteins.
Source: Ars Technica