Evo 2 AI is set to revolutionize synthetic biology, offering researchers a powerful tool to design genomes and predict mutations. Developed by the Arc Institute, NVIDIA, and top universities, the AI is trained on more than 128,000 genomes, making it an unprecedented force in bioengineering. Unlike traditional models, Evo 2 can handle long genetic sequences and even propose new CRISPR variants. It is open-source and integrated with NVIDIA’s BioNeMo framework, ensuring accessibility to scientists worldwide.

Building on its predecessor Evo 1, Evo 2 is trained on 9.3 trillion nucleotides from genomes spanning bacteria, archaea, phages, humans, plants, and other eukaryotic species. This makes it one of the largest AI models developed for biological research, capable of analyzing sequence lengths of up to one million nucleotides in a single pass. Its specialized architecture, StripedHyena 2, co-developed with OpenAI co-founder Greg Brockman, allows for the efficient processing of long sequences beyond what traditional transformer models can handle. Unlike conventional transformer-based models, which struggle with memory and computational limitations, Evo 2 leverages Fourier and convolutional techniques to extend its context window dramatically.

Evo 2’s capabilities go beyond simple genomic analysis. The AI has already demonstrated over 90% accuracy in identifying potentially pathogenic BRCA1 gene variants, distinguishing between benign and harmful mutations with precision. Additionally, Evo 2 can generate entire bacterial-scale genomes, including essential components like tRNA and rRNA genes. This functionality extends into creating entirely new biological mechanisms, such as designing transposons or genetic switches that activate only in specific cell types, a breakthrough that could improve gene therapy safety.

This breakthrough stands in contrast to AlphaFold, the AI developed by DeepMind, which has transformed protein structure prediction. While AlphaFold determines how proteins fold based on genetic information, Evo 2 operates at the genetic level itself, designing new sequences and modifying genomes for practical applications in medicine, agriculture, and biotechnology. Evo 2 examines and generates entire DNA or RNA sequences, sometimes spanning up to a million bases, while AlphaFold focuses on predicting individual protein structures. The two tools complement each other: Evo 2 can generate potential novel proteins or CRISPR systems, and AlphaFold can assess their likely 3D conformations.

The implications are enormous. Evo 2 accelerates bioengineering research, paving the way for personalized medicine, sustainable biotechnology, and synthetic organism design. Scientists could use Evo 2 to create bioengineered crops that resist disease and climate change, optimize microbes for industrial processes, or develop new forms of gene therapy. Researchers are even exploring the possibility of synthesizing an AI-designed genome in the lab, representing a proof-of-concept for fully AI-generated organisms.

Source.