Genomics · arcinstitute
opengenome2
A curated genomic database of 8.8 trillion base pairs across all domains of life for foundation model training.
What opengenome2 is
OpenGenome2 is a comprehensive genomic database compiled by the Arc Institute, containing approximately 8.8 trillion base pairs of curated DNA sequence data. It spans all primary domains of life, encompassing bacteria, archaea, eukaryota, and viruses. The collection was designed specifically to train and benchmark the Evo 2 family of genomic language models, but serves as a broader resource for large-scale biological foundation models and bioinformatics workflows.
The dataset integrates sequence data harvested from numerous public repositories. These include prokaryotic genomes from GTDB and IMG/PR, metagenomic assemblies from MGD DB, viral genomes from IMG/VR, eukaryotic sequences and annotations from NCBI and Ensembl, non-coding RNAs from RNAcentral and Rfam, promoter sequences from EPDnew, and various organellar genomes. A dedicated metadata file details the species and taxa represented across these sources.
How it was made
OpenGenome2 is organised into two distinct directory formats to support different computational pipelines: raw FASTA files containing biological DNA sequences, and preprocessed JSONL files tailored for model training. The JSONL format incorporates special tokens, context-specific formatting, and phylogenetic tags used during the pretraining of genomic models.
The data curation aligns with a two-stage training scheme. The first stage focuses on shorter sequences, combining prokaryotic genomes, metagenomic reads, stitched eukaryotic mRNAs, 5-kilobase genomic windows, organellar sequences, and non-coding RNAs. The second context-extension stage incorporates long-context sequences, extensive eukaryotic genomes across Animalia, Plantae, Fungi, Protista, and Chromista, and tagged prokaryotic and viral contigs.
Key facts
- Total size
- 8.8 trillion base pairs
- Available formats
- FASTA, JSONL
- Taxonomic coverage
- Bacteria, Archaea, Eukaryota, Viruses
- Primary application
- Training and benchmarking large-scale biological foundation models
- Accompanying metadata
- species_metadata.csv
What you can use it for
- Genomic Foundation Model Pretraining
- Train large sequence models across diverse taxonomic groups using the preprocessed JSONL files with phylogenetic tags and special tokens.
- Bioinformatics Sequence Analysis
- Perform comparative genomics, evolutionary conservation studies, and sequence searches using the raw FASTA assemblies across viral, prokaryotic, and eukaryotic domains.
- Non-Coding and Regulatory RNA Research
- Analyse structured non-coding RNAs and promoter elements curated from specialised resources including RNAcentral, Rfam, and EPDnew.
- Metagenomic Reference Benchmarking
- Use extensive metagenomic data from MGD DB alongside isolated bacterial and archaeal genomes for benchmarking metagenome assembly and binning tools.
Explore the data
The schema, real rows — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from arcinstitute/opengenome2 at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Provides expansive coverage spanning nearly 9 trillion base pairs across all domains of life and viruses.
- Supplies both raw FASTA sequences for general bioinformatics and preprocessed JSONL files for immediate machine learning use.
- Incorporates diverse genomic elements including full genomes, eukaryotic mRNAs, promoters, non-coding RNAs, and organellar DNA.
Watch out for
- The preprocessed JSONL files contain specialised tokens and phylogenetic tags structured specifically for the Evo 2 training regime, which may require adaptation for other model architectures.
- The vast size of the dataset demands substantial computational and storage resources to download and process.
Questions
Can I look inside opengenome2 without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser. Nothing is downloaded to your machine.
What licence is opengenome2 under?
Its authors declare apache-2.0. Read the terms on the original page before reusing the data.
What sequence formats are available in OpenGenome2?
The dataset is provided in two formats: FASTA for standard biological sequence analysis and JSONL containing preprocessed sequences with phylogenetic tags and special tokens for language model training.
Which public sources contributed to OpenGenome2?
Data were gathered from GTDBv220, IMG/PR, MGD DB, IMG/VR, NCBI, Ensembl, EPDnew, RNAcentral, Rfam, and diverse organellar records.
How are species and taxonomic classifications tracked?
A species_metadata.csv file is provided at the repository root detailing the taxa and source record accessions included in the collection.
Which organisms are covered in the eukaryotic portion?
The eukaryotic data include sequences from Animalia, Plantae, Fungi, Protista, and Chromista, incorporating both whole-genome long contigs and stitched mRNA and 5-kilobase genomic windows.
How to cite it
Brixi, G., Durrant, M. G., Ku, J., Poli, M., Brockman, G., Chang, D., Gonzalez, G. A., King, S. H., Li, D. B., Merchant, A. T., Naghipourfar, M., Nguyen, E., Ricci-Tam, C., Romero, D. W., Sun, G., Taghibakshi, A., Vorontsov, A., Yang, B., Deng, M., … Hie, B. (2025). Genome modeling and design across all domains of life with Evo 2. bioRxiv. https://doi.org/10.1101/2025.02.18.638918
Written from opengenome2's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.