Scientists from Nvidia and the Arc Institute, a Palo Alto-based nonprofit biomedical research organization, on February 19 unveiled an artificial intelligence model, Evo 2, that can not only identify disease-causing mutations in human genes, but also create new genomes.
Scientists from Nvidia and the Arc Institute, a Palo Alto-based nonprofit biomedical research organization, on February 19 unveiled an artificial intelligence model, Evo 2, that can not only identify disease-causing mutations in human genes, but also create new genomes.
The Evo 2 basic AI model, designed to gain a deeper understanding of biological code, was trained on the DNA of more than 100,000 species and life forms across various fields of biology, AIwire writes .
Evo 2’s predecessor, the Evo 1 model, was developed on single-cell genomes. The new model builds on the previous model but is trained on a larger amount of data. Evo 2 was trained using more than 9.3 trillion nucleotides — the building blocks of DNA or RNA — from more than 128,000 entire genomes, as well as metagenomic data. In addition to an expanded collection of bacterial, archaeal, and phage genomes, Evo 2 includes information on humans, plants, and other single-celled and multi-celled species from the eukaryotic realm.
The model was trained over several months on the NVIDIA DGX Cloud AI platform via AWS, using over 2,000 NVIDIA H100 GPUs with the support of the company’s researchers and engineers.
Patrick Hsu, co-founder of the Arc Institute, Arc Core Investigator, associate professor of bioengineering and Deb faculty fellow at the University of California, Berkeley, and co-author of the Evo 2 preprint, emphasized that the model enabled «machines to read, write, and think in the language of nucleotides.»
Evo 2, he said, has a general understanding of the tree of life that is effective in tasks such as predicting disease-causing mutations and developing potential code for artificial life. Scientists predict that the new AI model will be useful for genetic analysis and the development of new biological treatments.
Researchers from Stanford University, the University of California, Berkeley, and the University of California, San Francisco, in collaboration with colleagues at Nvidia, emphasize interpretation. The Evo 2 code has been made public on the Arc Institute GitHub and has also been integrated into Nvidia’s BioNeMo framework. The training data, code, and weights of the Evo 2 model, which the researchers are releasing, make it the largest fully open AI model of its kind, they say.
Evo 2 has already shown over 90% accuracy in determining which mutations in the BRCA1 gene (a gene associated with breast cancer) are benign or potentially pathogenic. This could potentially help reduce the cost and labor of experiments and streamline genetic research.
