With Evo 2, AI can model and design the genetic code for all domains of life

Evo 2 is a DNA foundation model published by Arc Institute and NVIDIA in the journal Nature, which can identify patterns in gene sequences across organisms from mammals to bacteria.
Trained on data from over 100,000 species, Evo 2 can predict disease-causing mutations in human genes and design new genomes comparable to simple bacteria.
Developed in collaboration with scientists from Stanford University, UC Berkeley, and UC San Francisco, Evo 2 is trained on over 9.3 trillion nucleotides from more than 128,000 genomes.
The model code is publicly available on GitHub, integrated into the NVIDIA BioNeMo framework, and includes tools for mechanistic interpretability visualization.
Evo 2 outperforms its predecessor Evo 1 by processing 30 times more data and reasoning over 8 times more nucleotides simultaneously, achieving high accuracy in genetic analysis.
Applications of Evo 2 include predicting Alzheimer’s risk and identifying genetic causes of various diseases, which may expedite the development of new medicines.
The research team has excluded potential pathogens harmful to humans from Evo 2 to mitigate ethics and safety risks.
The model is envisioned as a foundation for developing more specialized AI tools in genetics, potentially leading to new targeted treatments.

You may have missed