Large genome model: Open source AI trained on trillions of bases (arstechnica.com)

In a significant advancement for genomic research, the team behind the AI system Evo has announced Evo 2, a groundbreaking open-source AI model trained on trillions of base pairs from the genomes of bacteria, archaea, and eukaryotes. Unlike its predecessor, which was limited to bacterial genomes and their simpler structures, Evo 2 has demonstrated the ability to recognize complex genomic features in eukaryotes, including regulatory DNA and splice sites. This breakthrough is crucial as it enhances understanding of intricate genetic architectures that are often challenging for researchers to analyze manually. Evo 2's capability to develop internal representations of complex genome structures could revolutionize bioinformatics and genetic engineering by improving gene prediction and functional annotation. With eukaryotic genomes exhibiting interrupted coding sequences and loosely defined regulatory elements, Evo 2's advanced training allows it to navigate these complexities more effectively than previous models. This development not only paves the way for novel protein discovery but also invites further exploration of genomic functions across diverse life forms, positioning Evo 2 as a valuable tool for researchers seeking to unravel the intricacies of genetic information.