cshearer/LOL-EVE-TFBS-Benchmark

TFBS Benchmark Dataset

A benchmark dataset for evaluating variant effect prediction models on transcription factor binding site (TFBS) disruption tasks.

Dataset Description

This dataset contains 90,758 variants in transcription factor binding sites with comprehensive annotations for benchmarking variant effect prediction models, particularly the LOL-EVE model.

Dataset Summary

• Total variants: 90,758
• Species: Primarily human (homo_sapiens)
• Genes: Multiple genes with TFBS variants
• Transcription factors: Various TFs including CAMTA1, CAMTA2, CLOCK, E2F6, EBF3, ETV6, HAP1, HIC2, HIF1A, HSF1, KLF15, KLF7, MAX, MNT, and others
• Sequence context: 500bp promoter sequences centered on variants

Dataset Structure

Features

Feature	Type	Description
variant_id	string	Unique identifier for each variant
chromosome	string	Chromosome (e.g., chr1)
position	int64	Genomic position
ref	string	Reference allele
alt	string	Alternative allele
gene	string	Gene symbol
species	string	Species name (e.g., homo_sapiens)
tf	string	Transcription factor name
wt_seq	string	Wild-type sequence (500bp)
var_seq	string	Variant sequence (500bp)
mammalian_constraint	float32	Mammalian constraint score
expression_variability	float32	Expression variability score
distance_tss	int64	Distance to transcription start site

Data Splits

The dataset contains a single split with all 90,758 variants.

Usage

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("cshearer/LOL-EVE-TFBS-Benchmark")

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

# Example usage
example = train_data[0]
print(f"Variant ID: {example['variant_id']}")
print(f"Gene: {example['gene']}")
print(f"TF: {example['tf']}")
print(f"Position: {example['chromosome']}:{example['position']}")

Example Data

# Example variant
{
    'variant_id': 'chr1_1231946_CCGCCAACG_C',
    'chromosome': 'chr1',
    'position': 1231946,
    'ref': 'CCGCCAACG',
    'alt': 'C',
    'gene': 'b3galt6',
    'species': 'homo_sapiens',
    'tf': 'CAMTA1',
    'wt_seq': 'GAGCCAGACATCAAGGGCTCCACACAGCCGACTTCACATCTCCAAATCCTACTAACTGGGGATGAGGGTCCACGCGGTTCAGAAGCGGAAGCGCAGGCGCAGGGAAGCGGGGCAGCTTGTCCAAGGTCGCCTCGCCGATAAACGCGAGTCCAACCAGACCCCTTGGGCCTCCGTTTCCCGGTGGCATTCGTAGGTTTTGGCCAGTAGGAGACCAGACGTGCCGGCGGCCGGGGAGGCCAGCGTCGTCGGCCTGTCCCTGCCCCCGGGAACCCCGGGAGCCCCGGTGGCGGCGGAGTCTCGCCAGGGCTCAAGGCCGAGCGGACGGACGATGCCCCAGCCCAAGGCGGGAGGCGGCGGCGGCCTCCAGACCCGCCCTCGCCGTCCGGCCGGCGTACACTTGGCCCCGCGGCCTGCAGCGGCCGTCCCGGGCCCCTCACTCACCGGTCTGCCTCCCCGCGCTCGGGATCCGAGGACCGGAGCGAAGCGTCAGTGACGCCGCCAACGGGCCCGGATCAGGCCACTGCCATCTTTCTTGCGGGCGGGGGCGGTGCGAACGGGCGCGACCTCACGGAGGGGACGCCGGCGCCACCATCTCTCCTCCGGGCGGAAGCGGTCGCGGGGCCGCTCCGAGGTTGACCAATGACAAGGGTGCCCGAGGCCACGTGACGGCCGCCGATTGGCCGCCGGCCTCCGAGCGCCCCGGGGCTCGGCGTCTGCGGAAGGCCCCGGCGCGCTCCCAGGAGCGCCGTGCGCACGCGCACCGCCCCGAGCCGGCGGCGCCTGCGCA',
    'var_seq': 'GAGCCAGACATCAAGGGCTCCACACAGCCGACTTCACATCTCCAAATCCTACTAACTGGGGATGAGGGTCCACGCGGTTCAGAAGCGGAAGCGCAGGCGCAGGGAAGCGGGGCAGCTTGTCCAAGGTCGCCTCGCCGATAAACGCGAGTCCAACCAGACCCCTTGGGCCTCCGTTTCCCGGTGGCATTCGTAGGTTTTGGCCAGTAGGAGACCAGACGTGCCGGCGGCCGGGGAGGCCAGCGTCGTCGGCCTGTCCCTGCCCCCGGGAACCCCGGGAGCCCCGGTGGCGGCGGAGTCTCGCCAGGGCTCAAGGCCGAGCGGACGGACGATGCCCCAGCCCAAGGCGGGAGGCGGCGGCGGCCTCCAGACCCGCCCTCGCCGTCCGGCCGGCGTACACTTGGCCCCGCGGCCTGCAGCGGCCGTCCCGGGCCCCTCACTCACCGGTCTGCCTCCCCGCGCTCGGGATCCGAGGACCGGAGCGAAGCGTCAGTGACGCGGCCCGGATCAGGCCACTGCCATCTTTCTTGCGGGCGGGGGCGGTGCGAACGGGCGCGACCTCACGGAGGGGACGCCGGCGCCACCATCTCTCCTCCGGGCGGAAGCGGTCGCGGGGCCGCTCCGAGGTTGACCAATGACAAGGGTGCCCGAGGCCACGTGACGGCCGCCGATTGGCCGCCGGCCTCCGAGCGCCCCGGGGCTCGGCGTCTGCGGAAGGCCCCGGCGCGCTCCCAGGAGCGCCGTGCGCACGCGCACCGCCCCGAGCCGGCGGCGCCTGCGCACCTGCGCA',
    'mammalian_constraint': 0.9789,
    'expression_variability': 0.3259307772539296,
    'distance_tss': 708
}

Applications

This dataset is designed for:

1. Variant Effect Prediction: Benchmarking models that predict the functional impact of genetic variants
2. Transcription Factor Binding: Evaluating models that predict TF binding site disruption
3. Regulatory Genomics: Studying the impact of variants on gene regulation
4. Model Comparison: Comparing different variant effect prediction approaches

Data Collection

Source Data

The dataset is derived from variants in transcription factor binding sites with the following characteristics:

• Genomic regions: Promoter regions (500bp sequences)
• Variant types: Substitutions, insertions, and deletions
• Annotation sources: Mammalian constraint scores, expression variability, and TSS distance

Preprocessing

• Variants were filtered to include only those in TFBS regions
• Sequences were extracted as 500bp windows centered on variants
• Functional annotations were added from external databases
• Data was standardized and validated for consistency

Citation

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

@dataset{tfbs_benchmark_2024,
  title={TFBS Benchmark Dataset for Variant Effect Prediction},
  author={Marks Lab},
  year={2024},
  url={https://huggingface.co/datasets/cshearer/LOL-EVE-TFBS-Benchmark},
  license={MIT}
}

License

This dataset is released under the MIT License. See the LICENSE file for details.

Contact

For questions or issues related to this dataset, please contact the Marks Lab or open an issue on the dataset repository.

Related Datasets

• LOL-EVE Model
• eQTL Causal Dataset
• Ultra Rare Variants Dataset

Dataset Statistics

• Total variants: 90,758
• Unique genes: Multiple
• Unique TFs: 15+ transcription factors
• Sequence length: 500bp
• Species: Primarily human (homo_sapiens)
• Variant types: Substitutions, insertions, deletions
• Genomic coverage: Multiple chromosomes