Genomic Corpus · tattabio
OG
A curated prokaryotic and viral genomic corpus formatted with mixed amino acid and nucleotide sequences.
What OG is
Open Genomes (OG) is a curated genomic corpus developed by Tatta Bio for pretraining and benchmarking machine learning models in genomics. The resource is derived as a subset of the Open MetaGenomic dataset (OMG) and is approximately ten times smaller than the full collection.
The dataset focuses on high-quality prokaryotic and viral genomes paired with taxonomic details. To facilitate mixed-modality genomic language modelling, sequences are structured by translating protein-coding sequences into amino acids while keeping intergenic sequences in their raw nucleotide form.
How it was made
Each entry in the corpus corresponds to a genomic scaffold represented as an ordered progression of protein-coding sequences (CDS) and intergenic sequences (IGS). Coding regions are pre-processed and translated into amino acid sequences, whereas non-coding intervals are retained as nucleotide sequences.
Scaffolds contain parallel lists tracking sequence strings, their discrete positional indices along the scaffold, feature identifiers, and boolean flags denoting strand orientation for coding sequences. Identifiers record the sample accession, contig identifier, feature type, gene identifier, strand, and start-to-end coordinates.
Key facts
- Organism coverage
- Prokaryotes and viruses
- Sequence modalities
- Translated amino acids (CDS) and nucleotides (IGS)
- Size category
- Between 1 million and 10 million records
- Format
- Parquet
- Parent dataset
- Open MetaGenomic dataset (OMG)
What you can use it for
- Genomic language model pretraining
- Training mixed-modality transformer architectures that jointly process amino acid protein representations alongside non-coding nucleotide spans.
- Machine learning benchmarking
- Benchmarking genomic models on curated sets of prokaryotic and viral scaffolds containing taxonomic context.
- Intergenic and coding context analysis
- Evaluating structural and positional relationships between non-coding regulatory spaces and surrounding translated gene sequences.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from tattabio/OG at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 169,266 rows of default/train
CDS_ids
list- min
- 4
- median
- 19
- mean
- 62.29
- max
- 648
CDS_orientations
list- min
- 4
- median
- 19
- mean
- 62.29
- max
- 648
CDS_position_ids
list- min
- 4
- median
- 19
- mean
- 62.29
- max
- 648
CDS_seqs
list- min
- 4
- median
- 19
- mean
- 62.29
- max
- 648
IGS_ids
list- min
- 0
- median
- 16
- mean
- 51.85
- max
- 490
IGS_position_ids
list- min
- 0
- median
- 16
- mean
- 51.85
- max
- 490
IGS_seqs
list- min
- 0
- median
- 16
- mean
- 51.85
- max
- 490
Strengths and limits
Good at
- Integrates translated amino acids for coding segments and raw nucleotide sequences for intergenic intervals within uniform scaffold records.
- Provides curated prokaryotic and viral sequences supported by taxonomic annotations.
- Preserves positional ordering, genomic coordinates, and strandedness for all genomic features.
Watch out for
- Contains only prokaryotic and viral data, omitting eukaryotic genomes.
- Provides a substantially smaller sequence volume than the 3.1-trillion-base-pair full OMG corpus.
Questions
Can I look inside OG without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is OG under?
Its authors declare cc-by-sa-4.0. Read the terms on the original page before reusing the data.
How are the sequence modalities split across a scaffold?
Coding regions are provided as translated amino acid sequences, while intergenic regions between them are provided as nucleotide strings.
What information is encoded within the sequence identifiers?
Each identifier specifies the sample accession, contig identifier, feature type, gene identifier, strand direction, and start-to-end coordinate range.
How does OG differ from OMG and OMG_prot50?
OG is a taxonomically annotated subset roughly ten times smaller than the 3.1-trillion-base-pair OMG corpus, whereas OMG_prot50 is a protein-only corpus of 207 million sequences clustered at 50% identity.
How to cite it
Cornman, Andre, Jacob West-Roberts, Antonio Pedro Camargo, Simon Roux, Martin Beracochea, Milot Mirdita, Sergey Ovchinnikov, and Yunha Hwang. (2024). The OMG dataset: An Open MetaGenomic corpus for mixed-modality genomic language modeling. Cold Spring Harbor Laboratory. DOI: 10.1101/2024.08.14.607850.
Written from OG'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.