Update README.md
The OpenGenome2 was published as part of Genome modeling and design across all domains of life with Evo 2. Please refer to this paper for a more detailed description of the dataset and processing procedure.
Dataset Summary
OpenGenome2 is a massive, highly curated genomic atlas containing 8.84 trillion DNA base pairs, spanning all domains of life. It represents a significant expansion of the original OpenGenome dataset, increasing the total number of nucleotides from 300 billion to nearly 9 trillion.
This dataset was compiled to train Evo 2, a 40-billion parameter biological foundation model with a 1 million token context window. The dataset is designed to provide a comprehensive and representative view of genomic diversity, enabling models to learn the fundamental "language" of DNA. It includes prokaryotic, eukaryotic, metagenomic, and organelle sequences, with a specific focus on augmenting data around likely functional regions to improve performance on downstream biological tasks.
The primary purpose of OpenGenome2 is for the unsupervised pre-training of large-scale biological foundation models. The model trained on this data, Evo 2, is capable of:
- Zero-shot Variant Effect Prediction: Predicting the functional impacts of genetic variations without task-specific fine-tuning.
- Controllable Sequence Generation: Generating novel genomic sequences (mitochondrial, prokaryotic, eukaryotic) with specific desired properties.
- Fill-in-the-Middle (Masked Genome Modeling): Reconstructing missing segments of DNA sequences.
Source Data
The data was aggregated and curated from numerous public biological databases:
- Prokaryotic Genomes: Genome Taxonomy Database (GTDB) releases v214.1 and v220.0.
- Eukaryotic Genomes: NCBI Reference Sequences (RefSeq).
- Metagenomes: NCBI, JGI IMG, MGnify, MG-RAST, Tara Oceans samples, and others.
- Organelle Genomes: NCBI Organelle resource.
- Noncoding RNA: Ensembl, Rfam, and RNAcentral.
- Promoters: EPDnew database.
Extensive filtering, de-duplication, and quality control measures were applied, including clustering by average nucleotide identity (ANI) to select representative genomes, removing short or low-quality contigs, and filtering redundant metagenomic contigs based on protein clusters. The processing process is described in more details in Brixi et al., section 4.2.2..
Additional Information
Licensing Information
The dataset is released under the Apache License 2.0.
Citation Information
If you use this dataset in your research, please cite the following paper:
@article
{brixi2025genome,
title={Genome modeling and design across all domains of life with Evo 2},
author={Brixi, Garyk and Durrant, Matthew G and Ku, Jerome and Poli, Michael and Brockman, Greg and Chang, Daniel and Gonzalez, Gabriel A and King, Samuel H and Li, David B and Merchant, Aditi T and others},
journal={BioRxiv},
pages={2025--02},
year={2025},
publisher={Cold Spring Harbor Laboratory}
}