Evo 2 - A biological AI model developed by the Acr Institute in collaboration with NVIDIA, Stanford, and others.
Evo 2 is a DNA language model developed through collaboration between the Arc Research Institute, NVIDIA, Stanford University, and other institutions. It's used for genome modeling and design, covering all areas of life. Evo 2 is based on the StripedHyena 2 architecture...
What is Evo 2?
Evo 2 is a DNA language model developed through collaborations between the Arc Research Institute, NVIDIA, Stanford University, and other institutions. It's used for genome modeling and design, covering all areas of life. Based on the StripedHyena 2 architecture, Evo 2 handles context lengths of up to one million base pairs at single nucleotide resolution. It uses the OpenGenome2 dataset for autoregressive pre-training, which contains 8.8 trillion tags from all areas of life. Evo 2 supports long sequence modeling, DNA sequence generation, and embedding vector extraction, and provides multiple model checkpoints to meet diverse needs. Evo 2 advances genomics research and applications, providing powerful tools for fields such as biomedicine and synthetic biology.
Main functions of Evo 2
- Long context modelingIt can process DNA sequences of up to 1 million base pairs, supporting high-precision genome modeling.
- DNA sequence generationGenerates new DNA sequences based on given prompts, applicable to synthetic biology and gene editing.
- Embedded Vector Extraction: Extract the embedding vector of DNA sequence for downstream analysis, such as gene function prediction and variation effect analysis.
- Zero-sample predictionSupports zero-shot learning, such as predicting the impact of gene variations on function (e.g., predicting the effects of BRCA1 gene variations).
- Sequence scoring: Calculate the likelihood score of a DNA sequence to assess its stability and functional potential.
Evo 2's technical principles
- Large-scale data trainingIt was trained on data from over 9.3 trillion nucleotides, derived from over 128,000 genomes, covering organisms from multiple life forms including bacteria, archaea, and eukaryotes.
- Unique AI ArchitectureBased on the StripedHyena 2 architecture, it processes gene sequences of up to 1 million nucleotides to understand the relationships between distant parts of the genome.
- Deep Learning and Generative BiologyBased on deep learning technology, it aims to understand nucleic acid sequences in the same way it understands language. By learning biological sequence patterns formed during evolution, it predicts the impact of gene mutations and generates new genomes.
- Powerful computing supportThe training of Evo 2 utilizes NVIDIA's DGX Cloud AI platform and over 2,000 H100 GPUs, demonstrating powerful computing capabilities and efficient model training.
Evo 2 project address
- Project official website:https://arcinstitute.org/news/blog/evo2
- GitHub repository:https://github.com/ArcInstitute/evo2
- HuggingFace model library:https://huggingface.co/arcinstitute
- Technical Papers:https://arcinstitute.org/manuscripts/Evo2
Application scenarios of Evo 2
- Disease prediction: Identify whether gene mutations are pathogenic and assist in disease diagnosis.
- Gene therapyDesign cell-specific gene therapy tools to reduce side effects.
- Synthetic biologyDesigning new genomes to aid in artificial life research.
- Evolutionary researchTo identify gene sequence patterns and study biological evolution.
- Biotool developmentTo design tools such as biosensors to advance biotechnology.