BGI-HangzhouAI/Benchmark_Dataset-Genomic_element_classification

Summary

This dataset is constructed based on annotations from Ensembl Database. Through a filtration pipeline, genomic element sequences—including 3' UTR, 5' UTR, Exon, and Protein-coding gene—were extracted and integrated. The task aims to evaluate the model's ability to resolve complex, overlapping, and highly sequence-specific features within the genome.

Usage

from datasets import load_dataset
#Download the whole dataset
dataset = load_dataset("BGI-HangzhouAI/Benchmark_Dataset-Genomic_element_classification")
#Download a specific task
task_name = "Human_genomic_element_classification"
dataset = load_dataset(
    "BGI-HangzhouAI/Benchmark_Dataset-Genomic_element_classification",
    data_files = {
        "train": f"{task_name}/train.jsonl",
        "test": f"{task_name}/test.jsonl",
        "eval": f"{task_name}/eval.jsonl",
    }
)

Detailed Information

Task	Input fields	Train Seqs	Validation Seqs	Test Seqs	Max sequence length
Human_genomic_element_classification	{seq, label}	12,000	6,000	6,000	1,024
Mouse_genomic_element_classification	{seq, label}	12,000	6,000	6,000	1,024
Multi_species_exon_classification	{seq, label}	16,346	7,592	7,593	1,024

Human_genomic_element_classification / Mouse_genomic_element_classification

Genomic element	Label
3' UTR	0
5' UTR	1
Exon	2
Protein-coding gene	3

Multi_species_exon_classification

Species	Label
Human	0
Worm	1
Chicken	2
Fruit Fly	3
Mouse	4
Zebrafish	5

Final format

Datasets are saved in JSONL format. Each file contains:

• "seq" — the DNA sequence string (A/C/G/T, uppercase)
• "label" — multi class indicator