Marks-lab/LOL-EVE-eQTL_benchmark

LOL-EVE eQTL Causal Identification Dataset

Dataset Description

This dataset contains causal eQTL (expression quantitative trait loci) identification data for evaluating variant effect prediction models. The dataset includes indel variants in promoter regions with associated gene expression data and functional annotations.

Dataset Structure

• Total Variants: 7,090
• Species: Primates (Homo sapiens)
• Variant Types: Insertions and deletions in promoter regions
• Sequence Length: Up to 1,000bp promoter regions

Features

Basic Variant Information

• gene: Gene symbol
• species: Species name
• clade: Evolutionary clade
• chromosome: Chromosome
• position: Genomic position
• ref: Reference allele
• alt: Alternative allele
• variant_type: Type of variant (insertion/deletion/substitution)

Sequences

• wt_sequence: Wild-type sequence
• var_sequence: Variant sequence
• sequence_length: Length of sequence
• wt_sequence_start: Start position of sequence

Labels and Annotations

• pip: Posterior inclusion probability (causal label)
• indel_length: Length of indel
• distance_tss: Distance to transcription start site
• gsa_depth: GSA depth
• depth_category: Depth category

Functional Annotations

• phylop: PhyloP conservation score
• cadd: CADD score
• gc_percentage_delta: GC content change

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Marks-lab/LOL-EVE-eQTL")

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

# Example: Get causal variants (PIP >= 0.01)
causal_variants = dataset['train'].filter(lambda x: x['pip'] >= 0.01)
print(f"Causal variants: {len(causal_variants)}")

Citation

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

@article{loleve2025,
  title={A Genomic Language Model for Zero-Shot Prediction of Promoter
Variant Effects},
  author={Courtney A. Shearer, Rose Orenbuch, Felix Teufel,Christian J. Steinmetz, Daniel Ritter,
Erik Xie, Artem Gazizov, Aviv Spinner, Jonathan Frazer, Mafalda Dias,
Pascal Notin, Debora S. Marks},
  journal={Proceedings of the 19th Machine Learning in Computational Biology meeting},
  year={2025}
}

License

This dataset is released under the MIT License.