Large genome model: Open source AI trained on trillions of bases
System can identify genes, regulatory sequences, splice sites, and more.
TL;DR
- Evo 2, an open-source AI, is trained on trillions of base pairs from all three domains of life and viruses.
- It can identify key features in complex eukaryotic genomes, including regulatory DNA and splice sites, which were challenging for previous systems.
- The AI demonstrated proficiency in detecting mutations, recognizing RNA functions, and understanding different genetic codes across species.
- Evo 2 performed comparably to or better than specialized software in tasks like identifying splice sites and evaluating BRCA2 gene mutations.
- While capable of identifying features, its ability to design novel, functional proteins like its predecessor is less certain for eukaryotes.
- The AI model, training code, inference code, and dataset are publicly available for further research.