Overview

This dataset consists of reference genomes for 10 species:

Species	Assembly Accession
Arabidopsis thaliana	`GCF_000001735.4_TAIR10.1`
Caenorhabditis elegans	`GCF_000002985.6_WBcel235`
Danio rerio	`GCF_000002035.6_GRCz11`
Drosophila melanogaster	`GCF_000001215.4_Release_6_plus_ISO1_MT`
Felis catus	`GCF_018350175.1_F.catus_Fca126_mat1.0`
Gallus gallus	`GCF_016699485.2_bGalGal1.mat.broiler.GRCg7b`
Gorilla gorilla	`GCF_029281585.2_NHGRI_mGorGor1-v2.0_pri`
Homo sapiens	`GCF_000001405.40_GRCh38.p14`
Mus musculus	`GCF_000001635.27_GRCm39`
Salmo trutta	`GCF_901001165.1_fSalTru1.1`

Each item in the dataset contains the following fields:

"sequence": datasets.Value("string"),
"species_label": datasets.ClassLabel()  # See below
"description": datasets.Value("string"),
"start_pos": datasets.Value("int32"),
"end_pos": datasets.Value("int32"),
"fasta_url": datasets.Value("string")

The class labels are as follows:

Class	Label
`'Homo_sapiens'`	0
`'Mus_musculus'`	1
`'Drosophila_melanogaster'`	2
`'Danio_rerio'`	3
`'Caenorhabditis_elegans'`	4
`'Gallus_gallus'`	5
`'Gorilla_gorilla'`	6
`'Felis_catus'`	7
`'Salmo_trutta'`	8
`'Arabidopsis_thaliana'`	9

Usage

To use this dataset, set the chunk_length (length of each sequence, in base-pairs) and the overlap (amount each sequence overlaps, in base-pairs). The dataset only contains a train split. We recommend randomly splitting the dataset to create validation/test sets. See below for example usage:

import datasets

max_length = 32_768
overlap = 0

dataset = datasets.load_dataset(
  'yairschiff/ten_species',
  split='train',  # original dataset only has `train` split
  chunk_length=max_length,
  overlap=overlap,
  trust_remote_code=True
)
train_validation_splits = dataset.train_test_split(
  test_size=0.05, seed=42)

Acknowledgments

Code for dataset processing is derived from https://huggingface.co/datasets/InstaDeepAI/multi_species_genomes.