README.md

---
license: apache-2.0
dataset_info:
  features:
  - name: protein
    dtype: string
  - name: pdb_content
    dtype: string
  - name: file_size_bytes
    dtype: int64
  - name: protein_sequence
    dtype: string
  - name: mutant
    dtype: string
  - name: mutated_sequence
    dtype: string
  - name: dms_bin_score
    dtype:
      class_label:
        names:
          '0': Benign
          '1': Pathogenic
  - name: symbol
    dtype: string
  - name: mis_oe
    dtype: float64
  - name: af
    dtype: float64
  - name: ref_aa
    dtype: string
  - name: alt_aa
    dtype: string
  - name: aa_position
    dtype: int64
  splits:
  - name: train
    num_bytes: 37929504169
    num_examples: 62727
  download_size: 14538520642
  dataset_size: 37929504169
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
task_categories:
- text-classification
language:
- en
tags:
- biology
- medical
- genomics
- pdb
- protein-structures
- pathogenicity-prediction
- structural-bioinformatics
- ESMFold
- ProteinGym
pretty_name: Protein Structure Pathogenicity Dataset
size_categories:
- 10K<n<100K
---


# Protein Structure Pathogenicity Dataset

## Dataset Description

This dataset contains protein structures and metadata for benign and pathogenic missense variants, designed for training machine learning models to predict variant pathogenicity using protein structural information.

### Dataset Summary

The dataset includes:

- **Protein 3D structures** predicted via [ESMFold](https://github.com/facebookresearch/esm)
- **Benign and pathogenic variants** derived from the [ProteinGym](https://proteingym.org/) benchmark
- **Structural and sequence metadata** for each variant
- **Pre-computed features** including Allele Frequency and constraint metrics

This dataset was developed for the research paper:

> [**"Utilizing protein structure graph embeddings to predict the pathogenicity of missense variants"**](https://academic.oup.com/nargab/article/7/3/lqaf097/8211937) > _Authors: Martin Danner, Matthias Begemann, Miriam Elbracht, Ingo Kurth, and Jeremias Krause_

The dataset enables training of graph-based autoencoders to generate structural embeddings for downstream pathogenicity prediction tasks.

### Supported Tasks

- **Variant pathogenicity classification**: Binary classification of missense variants as benign or pathogenic
- **Protein structure analysis**: Analysis of 3D protein structures and their relationships to variant effects
- **Graph representation learning**: Training graph neural networks on protein structural graphs
- **Structural bioinformatics**: General structural analysis and feature extraction

## Dataset Structure

### Data Instances

Each instance in the dataset represents a single missense variant with its corresponding protein structure:

```python
{
    'protein': 'NP_000160.1',
    'mutant': 'T412I',
    'ref_aa': 'T',
    'alt_aa': 'I',
    'aa_position': 412,
    'dms_bin_score': 'Pathogenic',
    'pdb_content': '<PDB file content>',
    'protein_sequence': 'MQLRNPELHLGCALALRFLALV...',
    'mutated_sequence': 'MQLRNPELHLGCALALRFLALV...',
    'symbol': 'GLA',
    'mis_oe': 0.58230,
    'af': 0.000000,
    'file_size_bytes': 125847
}
```

### Data Fields

| Field              | Type   | Description                                                  |
| ------------------ | ------ | ------------------------------------------------------------ |
| `protein`          | string | RefSeq protein identifier (NP_XXXXXX.X format)               |
| `mutant`           | string | Amino acid substitution in standard notation (e.g., "T412I") |
| `ref_aa`           | string | Reference (wild-type) amino acid single-letter code          |
| `alt_aa`           | string | Alternate (mutant) amino acid single-letter code             |
| `aa_position`      | int    | Position of the mutation in the protein sequence             |
| `dms_bin_score`    | string | Binary pathogenicity label: "Benign" or "Pathogenic"         |
| `pdb_content`      | string | Complete PDB format structure file content                   |
| `protein_sequence` | string | Wild-type protein amino acid sequence                        |
| `mutated_sequence` | string | Mutant protein amino acid sequence                           |
| `symbol`           | string | Gene Symbol                                                  |
| `mis_oe`           | float  | Missense observed/expected ratio (constraint metric)         |
| `af`               | float  | Allele Frequency (0-1 scale)                                 |
| `file_size_bytes`  | int    | Size of the PDB structure file in bytes                      |

### Data Splits

Users should implement appropriate train/validation/test splits based on their specific use case.

### Dataset Statistics

- **Total variants**: ~64,000 missense variants

## Dataset Creation

### Source Data

#### Variants

The missense variants were derived from the [ProteinGym](https://proteingym.org/) deep mutational scanning (DMS) benchmark, which aggregates experimentally measured variant effects from multiple sources including:

- ClinVar
- gnomAD
- DMS experiments
- Clinical databases

#### Structures

Protein 3D structures were predicted using [ESMFold](https://github.com/facebookresearch/esm), a state-of-the-art protein structure prediction model based on protein language models. ESMFold generates accurate structural predictions directly from amino acid sequences.

## Considerations for Using the Data

### Limitations

- **Prediction quality**: Structures are predicted via ESMFold, not experimentally determined. Prediction confidence varies by protein.
- **Structural coverage**: Some proteins or regions may have lower-quality structural predictions.
- **Class imbalance**: The distribution of benign vs. pathogenic variants may not reflect natural prevalence.

### Recommended Use Cases

✅ **Appropriate uses:**

- Research on variant pathogenicity prediction methods
- Training and benchmarking ML models for structural biology
- Development of graph neural network architectures for proteins
- Educational purposes in computational biology

❌ **Not recommended:**

- Direct clinical decision-making without validation

## Citation

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

```bibtex
@article{10.1093/nargab/lqaf097,
    author = {Danner, Martin and Begemann, Matthias and Elbracht, Miriam and Kurth, Ingo and Krause, Jeremias},
    title = {Utilizing protein structure graph embeddings to predict the pathogenicity of missense variants},
    journal = {NAR Genomics and Bioinformatics},
    volume = {7},
    number = {3},
    pages = {lqaf097},
    year = {2025},
    month = {07},
    abstract = {Genetic variants can impact the structure of the corresponding protein, which can have detrimental effects on protein function. While the effect of protein-truncating variants is often easier to evaluate, most genetic variants that affect the protein-coding region of the human genome are missense variants. These variants are mostly single nucleotide variants, which result in the exchange of a single amino acid. The effect on protein function of these variants can be challenging to deduce. To aid the interpretation of missense variants, a variety of bioinformatic algorithms have been developed, yet current algorithms rarely directly use the protein structure as a feature to consider. We developed a machine learning workflow that utilizes the protein-language-model ESMFold to predict the protein structure of missense variants, which is subsequently embedded using graph autoencoders. The generated embeddings are used in a classifier model, which predicts pathogenicity. We provide evidence that graph embeddings can be used for pathogenicity prediction and that they can be used to enhance the widely applied CADD score. Additionally, we explored different levels of abstraction of the graph embeddings and their influence on the classifier. Finally, we compare the utility of graph embeddings from different protein-folding models.},
    issn = {2631-9268},
    doi = {10.1093/nargab/lqaf097},
    url = {https://doi.org/10.1093/nargab/lqaf097},
    eprint = {https://academic.oup.com/nargab/article-pdf/7/3/lqaf097/63841947/lqaf097.pdf},
}

```

### Related Resources

- **Code Repository**: [github.com/IHGGM-Aachen/genoseer](https://github.com/IHGGM-Aachen/genoseer)
- **ProteinGym Benchmark**: [proteingym.org](https://proteingym.org/)
- **ESMFold**: [github.com/facebookresearch/esm](https://github.com/facebookresearch/esm)

## License

This dataset is released under the **Apache 2.0** license.

- **Attribution**: You must give appropriate credit and indicate if changes were made

### Upstream Licenses

Please also respect the licenses of source data:

- **ProteinGym**: MIT
- **ESMFold predictions**: MIT

## Contact

For questions, issues, or feedback regarding this dataset:

- **GitHub Issues**: [github.com/IHGGM-Aachen/genoseer](https://github.com/IHGGM-Aachen/genoseer)
- **Email**: [mdanner@ukaachen.de](mailto:mdanner@ukaachen.de)

## Acknowledgments

We thank:

- The ProteinGym team for curating the variant benchmark
- Meta AI for developing and releasing ESMFold
- The gnomAD and ClinVar consortia for variant annotations
- The broader structural bioinformatics community

---

**Dataset Version**: 1.0
**Last Updated**: November 2024
**Maintained by**: Martin Danner and collaborators