andrewdalpino/Tiny-OpenGenome2

Tiny OG2 Dataset

This is a curated subset of the OpenGenome2 dataset consisting of over 1 million DNA samples with over 185 billion base pair (BP) tokens across 16 categories covering a broad spectrum of biological life. It is designed to replicate the distribution of samples used to train the Evo2 model but with substantially fewer training examples - making it ideal for knowledge distillation, rapid iteration, and academic use. It is divided into pretrain and midtrain subsets which are suited for short and long context training respectively.

Categories

Each pretrain and midtrain subset has a different set of categories.

Pretrain

The pretrain subset contains about 106B BP tokens divided over the following categories.

Category	Num Tokens	Weight	Comment
eukaryotic_genic_windows	90B	35%	5K BP stitched token windows.
gtdb_v220_imgpr	3.5B	18%	Genome Taxonomy Database + IMG/PR.
imgvr_untagged	468M	3%	IMG/VR viral sequences.
metagenomes	11B	24%	MGD database.
mrna	196M	9%	Eukaryotic mRNAs (Ensembl, NCBI).
mrna_splice_promoter	312M	9%	Stitched.
ncrna	17M	2%	RNAcentral, Rfam, Ensembl, NCBI.
organelle	422M	0.5%	Various organelles.
promoters	119K	0.02%	Eukaryotic Promoter Database new (EPDnew).

Midtrain

Midtrain contains roughly 80B BP tokens in long-context samples.

Category	Num Tokens	Weight	Comment
gtdb_v220_stitched	2B	13%	GTB tagged as long.
imgpr_long	18M	13%	IMG/PR samples tagged as long.
ncbi_genomes_animalia	43B	40%	Full genomes.
ncbi_genomes_chromista	630M	0.9%	Full genomes.
ncbi_genomes_fungi	3.6B	4%	Full genomes.
ncbi_genomes_plantae	29B	27%	Full genomes.
ncbi_genomes_protista	567M	0.9%	Full genomes.

Example Usage

Loading

To load the Tiny OpenGenome2 dataset using the HuggingFace Datasets library refer to the examples below.

First, install the datasets library using your favorite package manager.

pip install datasets

Then call the load_dataset() function, specifying the subset like in the examples below.

from datasets import load_dataset

# Load the pretrain subset.
dataset = load_dataset("andrewdalpino/Tiny-OpenGenome2", "pretrain")

# Load the midtrain subset.
dataset = load_dataset("andrewdalpino/Tiny-OpenGenome2", "midtrain")

Filtering

You can also filter the samples of the dataset like in the examples below.

dataset = dataset.filter(lambda sample: len(sample["sequence"]) <= 8192)

SELECTED_CATEGORIES = {
    "eukaryotic_genic_windows",
    "gtdb_v220_imgpr",
    "metagenomes",
}

dataset = dataset.filter(lambda sample: sample["category"] in SELECTED_CATEGORIES)

Code Repository

The code for this dataset can be found at https://github.com/andrewdalpino/TinyOG2.

References

• 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 Naghipourfar, Mohsen and Nguyen, Eric and Ricci-Tam, Chiara and Romero, David W and Sun, Gwanggyu and Taghibakshi, Ali and Vorontsov, Anton and Yang, Brandon and Deng, Myra and Gorton, Liv and Nguyen, Nam and Wang, Nicholas K and Adams, Etowah and Baccus, Stephen A and Dillmann, Steven and Ermon, Stefano and Guo, Daniel and Ilango, Rajesh and Janik, Ken and Lu, Amy X and Mehta, Reshma and Mofrad, Mohammad R.K. and Ng, Madelena Y and Pannu, Jaspreet and Re, Christopher and Schmok, Jonathan C and St. John, John and Sullivan, Jeremy and Zhu, Kevin and Zynda, Greg and Balsam, Daniel and Collison, Patrick and Costa, Anthony B. and Hernandez-Boussard, Tina and Ho, Eric and Liu, Ming-Yu and McGrath, Tom and Powell, Kimberly and Burke, Dave P. and Goodarzi, Hani and Hsu, Patrick D and Hie, Brian, Genome modeling and design across all domains of life with Evo 2, https://www.biorxiv.org/content/early/2025/02/21/2025.02.18.638918, 2025.
• GTDB (Genome Taxonomy Database): Parks, D. H., Chuvochina, M., Rinke, C., Mussig, A. J., Chaumeil, P.-A., & Hugenholtz, P. (2022). GTDB: an ongoing census of bacterial and archaeal diversity through a phylogenetically consistent, rank normalized and complete genome-based taxonomy. Nucleic Acids Research, 50(D1), D785–D794.
• Metagenomics (MGD DB): Durrant, M. G., Perry, N. T., Pai, J. J., Jangid, A. R., Athukoralage, J. S., Hiraizumi, M., McSpedon, J. P., Pawluk, A., Nishimura, H., Konermann, S., & Hsu, P. D. (2024). Bridge RNAs direct programmable recombination of target and donor DNA. Nature, 630(8018), 984–993.

Additional data sources include NCBI, Ensembl, IMG/VR, RNAcentral, Rfam, and EPDnew databases.