BacCorpus

BacCorpus is a large-scale bacterial genome corpus for training and evaluating genome-aware genomic language models. It contains quality-controlled bacterial genomes represented at the genome level, with predicted protein-coding sequences stored as protein translations and intergenic regions stored as DNA sequences.

Genomes were deduplicated at various sequence identity levels using genome sketching, specifically: 100, 99, 98, 95, 90, 85, 80. BacCorpus spans approximately 6.2 million unique genomes across more than 150,000 species, and over 10,000 environments. The dataset is intended for large-scale pretraining and analysis of bacterial genomic language models.

The table below shows the number of genomes at varying level of genome sketching sequence identity threshold. The lower sequence identity implies higher diversity:

The table below summarizes the number of genomes available at each genome-sketching sequence identity threshold. Lower sequence identity thresholds correspond to more aggressively deduplicated, more diverse genome sets. Configurations are cumulative: for example, the 95 configuration includes genomes from the 80, 85, 90, and 95 directories.

Configuration	Genome sketching identity threshold	Number of genomes
80	80%	92,257
85	85%	124,342
90	90%	159,152
95	95%	213,049
98	98%	634,467
99	99%	1,774,943
100	100%	6,196,770
all	unfiltered/all genomes	8,889,348

BacCorpus was built by combining genomes from several public bacterial genome resources, including MGnify [1], SPIRE [2], HRGM [3], GTDB [4], mOTUs DB [5], and AllTheBacteria [6]. Genomes were uniformly quality controlled and annotated.

Intended Uses

This dataset is intended for:

• pretraining bacterial genomic language models;
• studying bacterial genome organisation;
• building protein, DNA, mixed-modality, or genome-aware models;
• extracting protein or intergenic DNA sequences for representation learning;
• benchmarking model architectures that use genome context.

How to Use

Because BacCorpus is large, we recommend streaming the dataset rather than downloading it locally.

from datasets import load_dataset

ds = load_dataset(
    "AllTheBacteria/BacCorpus",
    name="80",  # choose one of: "80", "85", "90", "95", "98", "99", "100", "all"
    split="train",
    streaming=True,
)

example = next(iter(ds))
print(example.keys())

Schema

BacCorpus stores one row per genome. Each row contains aligned list-valued fields describing the features extracted from that genome.

Column	Type	Description
genome_id	string	Genome identifier.
contig_id	list<int64>	Contig index for each feature.
feature_id	list<int64>	Feature index within a contig.
molecule	list<string>	Feature type: AA for translated CDS, DNA for intergenic region.
start	list<int64>	Feature start coordinate.
end	list<int64>	Feature end coordinate.
strand	list<int64>	Feature strand, typically 1 or -1.
sequence	list<string>	Protein or DNA sequence for the feature.
source	string	Source database or resource.

The list-valued columns are aligned by position, so index i across contig_id, feature_id, molecule, start, end, strand, and sequence refers to the same genomic feature.

Other resources

For deduplicated and clustered DNA and protein sequence datasets, see:

• Protein sequences: https://huggingface.co/datasets/AllTheBacteria/BacCorpus-prot-90
• Intergenic regions (DNA): https://huggingface.co/datasets/AllTheBacteria/BacCorpus-intergenic-dna-90

Citation:

TBD

References

[1] Richardson, L. et al. MGnify: the microbiome sequence data analysis resource in 2023. Nucleic Acids Research 51(D1), D753–D759 (2023). doi:10.1093/nar/gkac1080

[2] Schmidt, T. S. B. et al. SPIRE: a Searchable, Planetary-scale mIcrobiome REsource. Nucleic Acids Research 52(D1), D777–D783 (2024). doi:10.1093/nar/gkad943

[3] Almeida, A. et al. A unified catalog of 204,938 reference genomes from the human gut microbiome. Nature Biotechnology 39, 105–114 (2021). doi:10.1038/s41587-020-0603-3

[4] Parks, D. H. et al. 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 (2022). doi:10.1093/nar/gkab776

[5] Dmitrijeva, M. et al. The mOTUs online database provides web-accessible genomic context to taxonomic profiling of microbial communities. Nucleic Acids Research 53(D1), D797–D805 (2025). doi:10.1093/nar/gkae1004

[6] Hunt, M. et al. AllTheBacteria – all bacterial genomes assembled, available, and searchable. bioRxiv (2024). doi:10.1101/2024.03.08.584059