Microbial Genomics · AllTheBacteria
Bac-Corpus-dna-intergenic-sequences-high-diversity
A clustered collection of 1.45 billion bacterial intergenic DNA sequences for pretraining genomic language models.
What Bac-Corpus-dna-intergenic-sequences-high-diversity is
BacCorpus Intergenic DNA 90 is an extensive dataset of representative bacterial intergenic DNA sequences. Intergenic regions are the non-coding stretches of DNA situated between protein-coding genes, functioning as reservoirs for regulatory and structural signals including promoters, terminators, and intergenic spacing.
The dataset is derived from BacCorpus, a bacterial genome resource encompassing millions of quality-controlled isolate genomes and metagenome-assembled genomes drawn from multiple public microbial repositories. By clustering sequences at 90% sequence identity and 90% coverage, it provides non-redundant representative sequences tailored for pretraining DNA and mixed-modality genomic language models.
How it was made
The corpus was constructed by collating quality-controlled bacterial genomes from six repositories: MGnify, SPIRE, HRGM, GTDB, mOTUs DB, and AllTheBacteria. Genomes were annotated, and the non-coding intergenic sequences were extracted. The extracted segments were subsequently clustered using MMSeqs2 (version 18.8cc5c) at criteria of 90% sequence identity and 90% coverage to reduce redundancy, yielding approximately 1.45 billion representative sequences formatted in Parquet.
Key facts
- Sequence count
- Approximately 1.45 billion representative sequences
- Clustering criteria
- 90% sequence identity and 90% coverage
- Clustering software
- MMSeqs2 (version 18.8cc5c)
- Primary source data
- BacCorpus (genomes from MGnify, SPIRE, HRGM, GTDB, mOTUs DB, and AllTheBacteria)
- Data format
- Parquet
- Biological domain
- Bacteria
- Sequence type
- Non-coding intergenic DNA
What you can use it for
- DNA language model pretraining
- Pretraining self-supervised foundation models on non-coding bacterial DNA to learn representations of regulatory architecture such as promoters and terminators.
- Mixed-modality genomic modelling
- Integrating intergenic regulatory sequences alongside protein coding sequences for comprehensive genome-aware machine learning models.
- Bacterial regulatory diversity analysis
- Investigating sequence variation and conservation across non-coding regulatory elements sampled broadly from bacterial diversity.
- Sequence redundancy benchmarking
- Evaluating the performance of sequence clustering and redundancy reduction strategies on large-scale microbial datasets.
Explore the data
No queryable preview
This dataset is not available in a queryable form — usually an unusual file format or a repository too large to convert. The original source may still be readable.Strengths and limits
Good at
- Provides approximately 1.45 billion representative intergenic sequences covering a broad spectrum of bacterial isolates and metagenome-assembled genomes.
- Reduces sequence redundancy systematically using MMSeqs2 clustering at 90% sequence identity and 90% coverage.
- Focuses explicitly on non-coding regulatory regions, addressing a component often under-represented in protein-centric genomic corpora.
Watch out for
- Contains only bacterial intergenic sequences, excluding eukaryotic and archaeal organisms as well as coding regions.
- Clustering at 90% identity collapses closely related variants, which removes fine-scale strain-level polymorphisms.
- The massive dataset scale requires data streaming rather than complete local downloads in resource-constrained environments.
Questions
Can I look inside Bac-Corpus-dna-intergenic-sequences-high-diversity without downloading it?
Not here: it is not available in a queryable form. Its description and statistics are still shown.
What licence is Bac-Corpus-dna-intergenic-sequences-high-diversity under?
Its authors declare apache-2.0. Read the terms on the original page before reusing the data.
What biological elements are present in these intergenic sequences?
These non-coding regions between bacterial genes typically encompass promoters, terminators, and spacing regions that regulate gene expression and genome organisation.
Which genomic resources contributed to the dataset?
Sequences were sourced from bacterial isolate genomes and metagenome-assembled genomes aggregated from MGnify, SPIRE, HRGM, GTDB, mOTUs DB, and AllTheBacteria.
How should this dataset be downloaded or accessed?
Because the dataset contains approximately 1.45 billion sequences, streaming the Parquet files via standard dataset libraries is recommended over downloading the full dataset locally.
Written from Bac-Corpus-dna-intergenic-sequences-high-diversity'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.