shmelev/genatator-segmentation-dataset

Ab inition gene segmentation benchmark (GENATATORs)

Overview

genatator-segmentation-dataset is a nucleotide-level gene segmentation dataset designed for training and evaluating DNA language models and related sequence models on transcript structure prediction. The dataset targets biologically detailed reconstruction of transcript architecture. In particular, it utilizes nucleotide-resolution labels describing the internal organization of transcripts, including 5' untranslated region (5' UTR), exon, intron, 3' untranslated region (3' UTR), and coding sequence (CDS).

The dataset was created for benchmarking and developing models for transcript segmentation in the context of ab initio gene annotation. It supports both human-only and multispecies training setups, as well as held-out validation and test evaluation.

Each sample contains exactly three fields:

• dna_sequence
• labels
• metadata

Intended use

This dataset is intended for:

• training DNA language models for transcript segmentation
• fine-tuning pretrained genomic foundation models
• benchmarking nucleotide-level and gene-structure-aware segmentation methods
• evaluating generalization from human-only to multispecies training
• studying segmentation of both protein-coding and long non-coding transcripts

It is particularly suitable for methods that operate on long genomic or transcript-derived sequences and output per-nucleotide labels.

Dataset configurations

The repository contains four configurations.

Config	Split	Description
train-human	train	Human-only training dataset. Sequence length is capped at 250 kb. One transcript per gene is retained, chosen as the isoform with the longest cumulative exon length.
train-multi-specie	train	Multispecies training dataset spanning human and additional mammalian assemblies. Sequence length is capped at 250 kb. One transcript per gene is retained, chosen as the isoform with the longest cumulative exon length.
val-human	validation	Human validation dataset. Sequence length is capped at 250 kb. One transcript per gene is retained, chosen as the isoform with the longest cumulative exon length.
test-human-complete	test	Human test dataset containing full transcript sequences from chromosome 20 of the T2T human genome, with no truncation by length and with all annotated transcripts of each gene retained.

The dataset follows different transcript selection rules for training, validation, and test.

• test-human-complete contains all transcripts of each gene from chromosome 20 of the T2T human genome.
• train-human, train-multi-specie, and val-human contain only one transcript per gene, selected as the transcript with the longest cumulative exon length.

Human chromosomes are partitioned so that chromosomes 8, 20, and 21 are held out from human training. In this setup:

• human training data exclude chromosomes 8, 20, and 21
• human validation data are drawn from held-out human chromosomes
• the full-length transcript version of human chromosome 20 is used for the test-human-complete dataset

For the multispecies dataset:

• all chromosomes are included for non-human species
• human examples follow the same held-out chromosome policy described above

Data schema

Each row has exactly three columns.

dna_sequence

A string containing the DNA sequence for the example.

• Type: string
• Alphabet: uppercase DNA characters (only A, T, C and G)

labels

A nested array of nucleotide-level target annotations aligned to dna_sequence.

• Type: nested numeric array
• Shape: sequence-length by class-dimension
• Interpretation: per-nucleotide multilabel or multiclass segmentation targets used for transcript structure prediction

The target class order is:

["5UTR", "exon", "intron", "3UTR", "CDS"]

metadata

A compact string encoding transcript-level annotation in the following format:

<type>|<gene_name>|<transcript_name>|<strand>|<start>:<end>|<genome>

Metadata fields

The metadata field contains biologically interpretable attributes packed into a single string. The meaning of each component is described below.

1. type

Transcript class.

Typical values include:

• mRNA
• lnc_RNA

This field indicates whether the transcript is protein-coding or long non-coding.

2. gene_name

Gene identifier or gene name associated with the transcript.

Typical values look like:

• gene-LOC124908100

This field identifies the parent gene for the transcript.

3. transcript_name

Transcript identifier.

Typical values look like:

• rna-XR_007089385.1
• transcript accession-like names from reference annotations

This field identifies the specific transcript isoform represented by the example.

4. strand

Genomic strand orientation.

Allowed values are typically:

• +
• -

This field indicates whether the transcript is encoded on the forward or reverse strand relative to the reference assembly.

5. start:end

Genomic coordinate interval associated with the example.

Example:

23090370:23092686

This field stores the coordinate span as:

• start: integer genomic start position
• end: integer genomic end position

6. genome

Genome or assembly identifier associated with the example.

Typical values are assembly accessions such as:

• GCF_009914755.1
• GCF_000001635.26

This field is especially relevant for the multispecies dataset, where it identifies the source assembly of the transcript example.

Multispecies training dataset

The train-multi-specie configuration includes data from 39 mammalian assemblies.

Assembly	Species
GCF_000952055.2	Aotus nancymaae
GCF_002263795.3	Bos taurus
GCF_000767855.1	Camelus bactrianus
GCF_000002285.3	Canis lupus familiaris
GCF_000151735.1	Cavia porcellus
GCF_001604975.1	Cebus imitator
GCF_000283155.1	Ceratotherium simum simum
GCF_000276665.1	Chinchilla lanigera
GCF_000260355.1	Condylura cristata
GCF_002940915.1	Desmodus rotundus
GCF_000151885.1	Dipodomys ordii
GCF_002288905.1	Enhydra lutris kenyon
GCF_000308155.1	Eptesicus fuscus
GCF_000002305.2	Equus caballus
GCF_018350175.1	Felis catus
GCF_000247695.1	Heterocephalus glaber
GCF_009914755.1	Homo sapiens
GCF_000236235.1	Ictidomys tridecemlineatus
GCF_000280705.1	Jaculus jaculus
GCF_000001905.1	Loxodonta africana
GCF_001458135.1	Marmota marmota
GCF_000165445.2	Microcebus murinus
GCF_000317375.1	Microtus ochrogaster
GCF_000001635.26	Mus musculus
GCF_900095145.1	Mus pahari
GCF_002201575.1	Neomonachus schauinslandi
GCF_000292845.1	Ochotona princeps
GCF_000260255.1	Octodon degus
GCF_000321225.1	Odobenus rosmarus divergens
GCF_009806435.1	Oryctolagus cuniculus
GCF_000181295.1	Otolemur garnettii
GCF_016772045.2	Ovis aries
GCF_000956105.1	Propithecus coquereli
GCF_003327715.1	Puma concolor
GCF_036323735.1	Rattus norvegicus
GCF_000235385.1	Saimiri boliviensis boliviensis
GCF_000181275.1	Sorex araneus
GCF_000003025.6	Sus scrofa
GCF_000243295.1	Trichechus manatus latirostris

Loading examples

Load a configuration from the Hugging Face datasets library:

from datasets import load_dataset

train_human = load_dataset("shmelev/genatator-segmentation-dataset", "train-human")["train"]
train_multi = load_dataset("shmelev/genatator-segmentation-dataset", "train-multi-specie")["train"]
val_human = load_dataset("shmelev/genatator-segmentation-dataset", "val-human")["validation"]
test_human = load_dataset("shmelev/genatator-segmentation-dataset", "test-human-complete")["test"]

Access one example:

sample = train_human[0]
print(sample["dna_sequence"])
print(sample["labels"])
print(sample["metadata"])

Summary

genatator-segmentation-dataset is a long-context, nucleotide-level transcript segmentation dataset for DNA language models and related genomic sequence models. It includes human-only and multispecies training resources, a human validation set, and a full-length human test set, all formatted for direct use with modern machine learning pipelines.