PromoterZoo

Dataset Description

This dataset contains promoter sequence training data for the LOL-EVE (Language of Life - Evolutionary Variant Effects) model. The dataset includes 13.6 million promoter sequences from various species, providing a comprehensive foundation for training genomic sequence models.

Dataset Structure

• Total Sequences: 13,611,373
• Species: Multiple species across different clades
• Sequence Length: 1,000 base pairs (promoter regions)
• Direction: Both towards and away from TSS

Features

• sequence: DNA sequence (1000bp promoter region)
• gene: Gene symbol
• species: Species name
• clade: Evolutionary clade
• chrom: Chromosome
• strand: Strand orientation (+/-)
• start: Genomic start position
• stop: Genomic stop position
• sequence_len: Length of sequence (1000)
• direction: Direction relative to TSS (towards/away)

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("cshearer/PromoterZoo")

# Access the data
print(f"Dataset size: {len(dataset['train'])}")
print(f"Features: {dataset['train'].features}")

# Example: Get sequences for a specific gene
gene_sequences = dataset['train'].filter(lambda x: x['gene'] == 'a1bg')
print(f"Sequences for a1bg: {len(gene_sequences)}")

# Example: Get sequences from a specific species
species_sequences = dataset['train'].filter(lambda x: 'homo' in x['species'].lower())
print(f"Human sequences: {len(species_sequences)}")

Data Processing

The sequences are processed as follows:

• Promoter regions are extracted 1000bp upstream of transcription start sites
• Both forward and reverse strand sequences are included
• Sequences are labeled with gene, species, and genomic coordinates

Citation

If you use this dataset in your research, please cite:

@article{loleve2024,
  title={LOL-EVE: Language of Life - Evolutionary Variant Effects},
  author={Your Name and Collaborators},
  journal={Nature},
  year={2024}
}

License

This dataset is released under the MIT License.