YYJMAY/pm-binding

PM (Peptide-MHC) Binding Prediction Dataset

Dataset Description

This dataset is part of the SPRINT benchmark framework for TCR-pMHC binding prediction. It contains peptide-MHC binding data for training and in-distribution testing of binding prediction models.

Dataset Summary

The PM dataset focuses on peptide-MHC binding prediction without TCR information. It is reorganized and standardized from multiple sources to provide a clean benchmark for PM task evaluation.

Dataset Structure

Files

• train.csv: Training data
• id_test.csv: In-distribution test data

Data Format

Each CSV file contains the following columns:

Column	Type	Description
peptide	string	Peptide amino acid sequence
HLA	string	HLA allele (standardized format: A*02:01)
label	int	Binary binding label (0=non-binder, 1=binder)
length	int	Peptide length (8-14 amino acids)
HLA_sequence	string	HLA pseudo-sequence

Dataset Statistics

Training Set:

• Total samples: 1683280
• Label type: Binary (0/1)
• Positive rate: 18.25% (307201/1683280)
• Unique HLAs: 112
• Unique peptides: 1481879
• Peptide length range: 8-14

Test Set (In-Distribution):

• Total samples: 586608
• Label type: Binary (0/1)
• Positive rate: 14.64% (85876/586608)
• Unique HLAs: 112
• Unique peptides: 558035
• Peptide length range: 8-14

Usage

Load with Pandas

from huggingface_hub import hf_hub_download
import pandas as pd

# Download files
train_file = hf_hub_download(
    repo_id="YYJMAY/pm-binding",
    filename="train.csv",
    repo_type="dataset"
)
test_file = hf_hub_download(
    repo_id="YYJMAY/pm-binding",
    filename="id_test.csv",
    repo_type="dataset"
)

# Load data
train_df = pd.read_csv(train_file)
test_df = pd.read_csv(test_file)

Use with SPRINT Framework

from sprint.core.dataset_manager import DatasetManager

manager = DatasetManager()
config = {
    'hf_repo': 'YYJMAY/pm-binding',
    'files': ['train.csv', 'id_test.csv']
}

files = manager.get_dataset('pm', config)

Data Preparation

This dataset was prepared with the following steps:

1. Source Integration: Combined data from multiple PM binding datasets
2. HLA Standardization: Normalized HLA allele names to A*02:01 format
3. Quality Control: Removed duplicates and incomplete entries
4. Column Standardization: Unified column names and formats
5. Validation: Checked for data consistency and quality

Tasks

This dataset is designed for:

• Peptide-MHC Binding Prediction: Predicting binding affinity between peptides and MHC molecules
• In-Distribution Evaluation: Testing model performance on similar data distribution as training
• Baseline Comparison: Standardized data for reproducible benchmarking

Citation

If you use this dataset, please cite:

@dataset{pm_dataset_2024,
  title={PM (Peptide-MHC) Binding Prediction Dataset},
  author={SPRINT Framework Contributors},
  year={2024},
  url={https://huggingface.co/datasets/YYJMAY/pm-binding}
}

License

MIT License

Contact

For questions or issues, please refer to the SPRINT framework repository.