Scientists at the Arc Research Institute introduce Evo, an AI model trained on 2.7M microbial genomes that can both interpret and generate genetic sequences with unprecedented accuracy.
Unlike traditional language models trained on text, Evo simultaneously learns from DNA, RNA, and protein sequences. Evo has already designed working genetic editing tools and can generate entirely new genome-length sequences over 1M base pairs long, though they aren’t capable of forming fully viable organisms yet. The company has also accurately predicted how DNA changes would affect bacteria.
Evo has deliberately excluded human-affecting viral genomes from training for safety reasons.
Having a ‘ChatGPT for DNA’ has the potential to drastically accelerate research, design new proteins for drug development, predict disease-causing mutations, and more. However, it also raises questions about how society will responsibly handle tools that can write genetic code as easily as we write emails.
SOURCE MATERIAL : Sequence modeling and design from molecular to genome scale with Evo. SCIENCE. 15 Nov 2024 Vol 386, Issue 6723. DOI: 10.1126/science.ado9336 https://www.science.org/doi/10.1126/science.ado9336
