Dataset Viewer
Auto-converted to Parquet Duplicate
genome_id
stringlengths
9
17
contig_id
stringlengths
22
31
dna
large_stringlengths
501
2.99M
contig_length_bp
int64
501
2.99M
num_genes
int64
0
2.84k
genes
listlengths
0
2.84k
domain
stringclasses
2 values
phylum
stringclasses
57 values
class
stringclasses
141 values
order
stringclasses
357 values
family
stringclasses
722 values
genus
stringlengths
6
19
species
stringlengths
17
37
SPIREOTU_02369511
SPIREOTU_02369511__000000000001
CAATGAGGCCATTAGGATGGCTAAGTCCCAACGAATTTTTAAGAAAATATAAAAGTCAAGGGAAATCGGTCGTAACAATATAGAGCACTCAAACTACCTATTTTTCAAAAAAGTGTGACACATGATTGACTAACCTACAAGTAATTTTTATAAAAATCACAAAAACTATGTACAAAAAAACATATTTAATGATATAATTCATTATGAGAAAATGACTAAAGAGAGAAAGGAAAAAACTTATGGGAAAAATTGTGTTATTAGATGACAAAACAATAAATAAAATTGCAGCTGGTGAAGTTATTGAAAGACCTGCTTCTGTA...
25,401
20
[ { "gene_id": 0, "start_bp": 239, "end_bp": 2525, "strand": "+", "cog_id": "COG0323", "protein": "MGKIVLLDDKTINKIAAGEVIERPASVVKEIVENSIDAGATNISVEIQNGGISYIRVTDNGKGIMQDDMEIAFERHATSKLRSAEDLNDIKSMGFRGEALASIAAIAKVDLISKTADSDNGYKVTIEGGIILDKEETACQNGTTITIENLFFNTPVRYKFLKKDFTESGYIEDAITHVALVHPEIAIKLINTGKT...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000002
AAAAAATGAAAGAGGAGGGCGTAGAATATGGAAGCTTAGGTTTAGCTCCTCTTGCTAATATAATAAACGAAGATAAAAAGCCAACTGTGGTTGAACGAGTACTAGATTATATTTATGAGAACATGAATAAGTTTTATGGCTTCAAGACACTTTATCATTATAAAAAAGACTATGGACCAACGGATTGGGTAACTAGATACTTGATATTTTATCCTAAAACGTTTACTCCTAAAATAGCATATTCAATTATAAAAGCACAAAATCCTAAGGGAGTTAAAGATTATATATTTACAGCAATTAAAAAAATATTTTTAAAGGAT...
2,515
2
[ { "gene_id": 361, "start_bp": 2, "end_bp": 344, "strand": "+", "cog_id": "COG2898", "protein": "KMKEEGVEYGSLGLAPLANIINEDKKPTVVERVLDYIYENMNKFYGFKTLYHYKKDYGPTDWVTRYLIFYPKTFTPKIAYSIIKAQNPKGVKDYIFTAIKKIFLKDNFIKDNN", "partial": true }, { "gene_id": 362, "start_bp": 486, "end_b...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000003
TTTTTGTATGATATTATTACTCATAAGAGAGTTCCTCCTGATATGTTTATTTTTGGTCACTTAAACAATATCATTTTTGGTACTCTCTTTCAATATTTTATTTTCAATTTGCATGTGACATATCTATTTTAAACCTATAAAAATGAAAAATTCAAAAATTTATTTGTAAAAAAATAATATAGTTATCTTAGATTCTTTTTATAGAAAACGTTCACATAACTTTCACATAAAAAATTGACAATCTCTATTGACAATCACCAATAAAGGATGTATTATATATAAAGTTTAGCACTCTAAACCTTAAAGTGCTAAAAAGAGGT...
5,741
5
[ { "gene_id": 615, "start_bp": 326, "end_bp": 1385, "strand": "+", "cog_id": "COG1420", "protein": "MLDDRKKKILQAIIEEYIETAEPVSSKAIMGKRSIDYSSATIRNEMAELEKVGLLDKPHTSSGRVPSAKGYRYYVDELLNYNDISMQEIKYIQAQLETKVNQIEDLTKITTSTLSEITHYTSVGIGPRVDSEDIQEIKFILLGSRLLMAVILTDNGLIKETIIKFDDDITEEQVETLTYLFNNRLKGKPLNEI...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000004
AATGAGGCCATTAGGATGGCTAAGTCCCAACGAATTTTTAAGAAAATATAAAAGTCAAGGGAAATCGGTCGTAACAATATAGAGCACTCAAACTACCTATTTTTCAAAAAAGTGTGACACATGATTGACTAACCTACATTAAAACTGAAAATTATTGTCAAAATCTATTGACAATAATTTTTAGTATGTTAAAATAATAAAAGGAAAAATATATTTAAAGGAGAGAAACTATGGGAGAAGATAATAATAGACAATTAAAAAACGAAAGATTACAAAGTCAAGTAAACGAATTAAATGAATGTAATGCTGAATTGAACAGC...
8,765
10
[ { "gene_id": 962, "start_bp": 230, "end_bp": 371, "strand": "+", "cog_id": null, "protein": "MGEDNNRQLKNERLQSQVNELNECNAELNSLTKQYQELRQQETRAI", "partial": false }, { "gene_id": 963, "start_bp": 398, "end_bp": 779, "strand": "+", "cog_id": null, "protein": "MQEVT...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000005
AAAAAGGATATAAAGATGGAAAAGGCGGTAGAGGAGATTTGATTTTACAAGTAAAAATAATGATTCAAAGAAACATATCAGACGGAGAAAAAGAGCTATATAAACAACTTAAAGAAATTTCAAAATTTAATCCAAGACAGCAGTACAACTAAAAACATAAAAAGTAGACATAAATATATTGACATATTTATCCAGGTATAGTATAATAAAAGTGTGGAGAACACATGTTATACTATACAAAAGACTAAAATATGAAATCATATTTCATATGGAAAGAGAGGAAATAAAAAATGGACAATATACAAGGTAAACACTTTAGT...
5,465
7
[ { "gene_id": 20, "start_bp": 2, "end_bp": 152, "strand": "+", "cog_id": "COG0484", "protein": "KGYKDGKGGRGDLILQVKIMIQRNISDGEKELYKQLKEISKFNPRQQYN", "partial": true }, { "gene_id": 21, "start_bp": 290, "end_bp": 818, "strand": "+", "cog_id": null, "protein": "MD...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000006
ACGAAGCTATAAAAGCACCAGATCCATATGCAATTGTAAATTCAATTGATAGTACAGGTCATGGGTCACATGTTGTTGGAATTGCGTGCGGAGGAGGAAAAATAAATTTAAAATATAAATAATTAATTTAAAATGTAAAAAAATAAAATTAATTAAATAATTTAACAAAAATAATTTTAATTAATTATTTAAATAATATAAAACTAATTATAAATAAAATTAAATAAAACAATCGAAACATTAAAAAAATAAATAAATAAAAAGTAATTTAATTATTTATTTAAATTACTTTTTATTTATGTTTATTTATTTTTAAATCT...
6,352
9
[ { "gene_id": 363, "start_bp": 2, "end_bp": 122, "strand": "+", "cog_id": "COG1404", "protein": "EAIKAPDPYAIVNSIDSTGHGSHVVGIACGGGKINLKYK", "partial": true }, { "gene_id": 364, "start_bp": 429, "end_bp": 600, "strand": "+", "cog_id": null, "protein": "MVFVILFFLG...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000007
ACAATTGTAACATGGGACAAGTATAAAGATGAGTTAATAAAAGAGAACATAATAACTGGTGATGATGATGTAGAAGATATAGAAAATGGTAAAAAGCAAATAGTAACAGATACAGGATATGTAGCAGTGATAGAAATTGTACCAGATGGAACTGTAGGAAAAATAGAGATAATAGGTAAGAAAGATAGCTTAAAAGTGAATATAATAAATGTGGCAGTGACATCTGAAACAAGTAGCATAAAAGTAGATGTAACAGCAAAGAGAACAGACAATGCAACATTTAAATATTACTATAAAATAGAAGGTGGAGAATATCAAAC...
37,329
27
[ { "gene_id": 620, "start_bp": 0, "end_bp": 1923, "strand": "+", "cog_id": null, "protein": "TIVTWDKYKDELIKENIITGDDDVEDIENGKKQIVTDTGYVAVIEIVPDGTVGKIEIIGKKDSLKVNIINVAVTSETSSIKVDVTAKRTDNATFKYYYKIEGGEYQTSPVHEGTDLSYTITNLAQDTTYTIKVEATNKNGTVEKEAVARTGEVAPAQGAITFDNLTWTSGKASVKVNKAISDNLQIQYKVNTGSYAIITS...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000008
CTAATACTAATGTTAGTGTTATTCCTGCTAATATTATTAGTACTACTATTGTTACAACTAAAGCAACTAATGTGATTCCTTTTTTATTCTTTATTCCTTCATTTTTCATCTTTTTTCTCCTTTTTTCTTCAGTCATTTTTAATTTTAGTTTTATTCTTATTGTTTTTTACTCTCTTTGTTTAAATATTCTGTATAGTTTTTTAAAAGACTTATATACTTTCTTCTTAATTAGTGCTTTATTATATTTATTTTAGCTTTTGCAGTACTCAGCACTACATATATAAAGCAATCCTAAAAGCAACTGTGCTATCAGAAAAGGT...
5,869
7
[ { "gene_id": 972, "start_bp": 1, "end_bp": 109, "strand": "-", "cog_id": null, "protein": "MKNEGIKNKKGITLVALVVTIVVLIILAGITLTLVL", "partial": true }, { "gene_id": 973, "start_bp": 272, "end_bp": 1412, "strand": "-", "cog_id": null, "protein": "MKKIKNQEQKDEKGITL...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000009
ATTTTTGTATGATATTATTACTCATAAGAGAGTTCCTCCTGATATGTTTATTTTTGGTCACTTAAACAATATCATTTTTGGTACTCTCTTTCAATATTTTATTTTCAATTTGCATGTGACATATCTATTTTAAACCTATATCAATTAAACATTTTTAAAATTTCTGTTTTATCTTGCACTCCAACTAATGTATTAACTACTTTTCCGTTTTTAATTATTACAAATGTAGGTATGCTCATTACATCAAATTTTGATGCTAAATCTCTTTCTTGGTCTACATCGGCTTTGCATACTTTTATCCTTCCAACATTTTCATTTGC...
14,381
15
[ { "gene_id": 27, "start_bp": 140, "end_bp": 452, "strand": "-", "cog_id": "COG3118", "protein": "MNEVILTNKNYEKEVEKSEGPVLIDFWATWCGPCRSMMPVVEEIANENVGRIKVCKADVDQERDLASKFDVMSIPTFVIIKNGKVVNTLVGVQDKTEILKMFN", "partial": false }, { "gene_id": 28, "start_bp": 515, "end_bp": 1391,...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
SPIREOTU_02369511
SPIREOTU_02369511__000000000010
AAATTATGTAGTAGTTATGCATGGGATACAGCAATAAACTTTATACAAAAGACAGTAGCAAATTATGGAACAAGTTCAAATCAAGGAAATTATAATGATACAACATTCACATATACAGATATAACGGGAGCAACACAAACAAAAGCATCAAGTAAAAGTGTATTAATACCAACAGGACAAACAACACCAATAAATAGCATATATGATATGGGAGGAAATGACTGGGAGTGGACAACAGAGGCATCTAGCGTTTCAGGTACTCCTTATGTCTTTCGTGGAGGCAGTTGCGGCACTACATACTCCGAAGTTCCTGCTGGGTA...
3,363
3
[ { "gene_id": 372, "start_bp": 0, "end_bp": 387, "strand": "+", "cog_id": null, "protein": "KLCSSYAWDTAINFIQKTVANYGTSSNQGNYNDTTFTYTDITGATQTKASSKSVLIPTGQTTPINSIYDMGGNDWEWTTEASSVSGTPYVFRGGSCGTTYSEVPAGYRYCYTTTISDSVIALRATLYM", "partial": true }, { "gene_id": 373, "start_bp": 735, ...
Bacteria
Bacillota
Clostridia
TANB77
CAG-508
SPIREOTU_02369511
SPIREOTU_02369511 SPIREOTU_02369511
End of preview. Expand in Data Studio

Coarse-to-Fine Genomes

Genomic data from GLOBDB release 232, provided as gene-order sequences, protein sequences and DNA training examples.

How we selected the genomes

We started with the 346,233 microbial genomes in GLOBDB r232, which combines GTDB r232 with additional genome collections. We required CheckM2 completeness of at least 90%, contamination below 5%, and an assigned genus.

For each genus, we selected the genome with the highest contig N50, to favor less fragmented assemblies. Ties were resolved by higher completeness, then lower contamination, then alphabetical genome ID. This gave 45,365 genomes, one per genus.

We randomly selected 1% of these genera for validation, using seed 42 and rounding up. This gave 454 validation genera. Every genome from those genera is excluded from training. Whole genomes stay in one split, so contigs from the same genome cannot appear in both.

Which genomes each dataset uses

Dataset Selection Training genomes Validation genomes
contigs One genome per genus 44,911 454
genes All eligible genomes from training genera 183,523 454
proteins One genome per genus 44,911 454
dna One genome per family 8,004 74
flat_dna One genome per genus 44,911 454

The protein data uses the one-per-genus selection. For the DNA training data, we choose one of these genomes per family, using the same N50 and quality ranking. Each selected genome keeps its existing train/validation assignment.

For gene-order training, we use all genomes that passed the quality filter and belong to training genera. Gene validation uses the same genus representatives as protein validation. One genome, QXLSG_01327, was excluded because its source database contained duplicated contigs and disagreed with its reported size.

genomes.parquet lists the selected genomes, their taxonomy, quality scores, split and membership in each dataset.

How we prepared the training examples

The following steps apply to genes, proteins and dna.

  • Gene labels: use the first supplied COG24 assignment. Keep genes without a COG as UNASSIGNED. Gene tokens include strand, such as COG0123+ or COG0123-.
  • Gene quality: remove partial genes, ambiguous sequences and genes whose DNA does not translate to the supplied protein. Translation uses the genome's CheckM2 genetic code. If one annotation lies entirely inside another, keep the outer gene. If a gene fails filtering, remove its group of overlapping genes.
  • Sequence boundaries: keep genes on either side of a rejected region or an ambiguous intergenic gap in separate examples. Never join separate contigs.
  • Length limit: use at most 8,192 tokens, including formatting. Split gene sequences between genes, protein sequences between complete proteins, and DNA between groups of overlapping genes. Skip a unit from a table if it cannot fit intact. Proteins run N-to-C and include a terminal stop token in their length.

The DNA table puts coding sequences first and then generates their intervening gaps with fill-in-the-middle (FIM) formatting:

coding A <gap> coding B <gap> coding C <fill>
gap 1 DNA <end_gap> gap 2 DNA <end_gap> <eos>

Shared bases in overlapping genes occur once. Gaps at contig ends or example boundaries are omitted. The contigs data preserves those original sequences.

Load the data

from datasets import load_dataset

data = load_dataset(
    "Iddah/coarse-to-fine-genomes", "contigs", split="train", streaming=True
)
example = next(iter(data))

Choose a dataset name from the table above. Every dataset has train and validation splits. For a DNA-only baseline, load flat_dna. Token IDs, taxonomy mappings and exact training-table counts are in metadata.json. The prepared tables contain unpacked examples; packing is left to the training loader.

Original contigs

contigs contains the full DNA sequences, proteins, coding-gene coordinates, COG labels and taxonomy for the one-per-genus selection. It is 65.24 GB compressed, with 135.44B training bases and 1.35B validation bases.

Each row is one complete contig. Its genes list gives each coding gene's ID, start, end, strand, protein sequence, COG ID and partial-gene flag. Coordinates are 0-based, half-open. Gene IDs are unique within a genome. Missing COGs are null. These source records retain partial genes and ambiguous sequences as supplied.

Flat DNA baseline

flat_dna contains plain DNA windows from the same one-per-genus contigs, preserving their train/validation assignment. We split at every non-ACGT character, uppercase the remaining DNA, and divide each continuous interval into windows of at most 8,192 bases. Short contigs and final fragments are retained. Windows can cut through genes but never cross contig boundaries or ambiguous bases.

Split Windows DNA bases
train 20,007,405 135,392,160,275
validation 221,190 1,350,242,438

The files total 41.53 GB compressed. Each row contains genome_id, contig_id, start_bp, end_bp and dna. Coordinates are 0-based, half-open in the original contig. DNA is stored as strings, without special tokens or FIM formatting. Tokenization and packing are left to the training loader; attention must stay within each independent sequence.

Citation and license

This dataset was selected, filtered and reformatted from GLOBDB r232. It is released under CC BY-SA 4.0, following GLOBDB. Please cite Speth et al., GlobDB and the relevant source datasets and annotation tools.

Downloads last month
79