---
readme: "true"
language:
  - en
tags:
  - anti-CRISPR # 如果是文本就改成 text-classification
  - benchmark
  - Acr
task_categories:
  - text-classification
configs:
  - config_name: default
    data_files:
      - split: positive
        path: "positive.tar.gz"
      - split: negative
        path: "negative.tar.gz"
---

# ShieldBreaker Benchmark Dataset

## Overview

The ShieldBreaker Benchmark Dataset is a comprehensive collection of anti-CRISPR protein sequences and structures, designed for machine learning research in CRISPR-Cas system inhibition. This dataset contains both positive (anti-CRISPR) and negative (non-anti-CRISPR) samples with dual-modal data representations.

## Dataset Structure

```
ShieldBreaker_Upload/
├── positive/
│   ├── fasta/
│   │   └── benchmark_positive_filtered.fasta
│   └── pdb/
│       └── [1,421 PDB structure files]
├── negative/
│   ├── fasta/
│   │   └── negative_changename_filtered.fasta
│   └── pdb/
│       └── [52,824 PDB structure files]
└── README.md
```

## Data Description

### Positive Samples

Anti-CRISPR proteins that actively inhibit CRISPR-Cas systems:

- **FASTA format**: 1,421 protein sequences with standardized naming convention
- **PDB format**: 1,421 corresponding 3D protein structures
- **Naming**: `anti_CRISPR{ID}_{Accession}_{AcrFamily}_{Genus}_label_1` or `new_{ID}_{Accession}_{AcrFamily}_{Genus}_label_1`
Among these, the naming format starting with new_ refers to newly collected samples based on antiCRISPRdb 2.2

### Negative Samples

Non-anti-CRISPR proteins used as negative controls:

- **FASTA format**: 52,824 protein sequences (105,648 lines total)
- **PDB format**: 52,824 corresponding 3D protein structures
- **Naming**: Standardized protein identifiers with label information
- **Source**: Carefully selected from all prokaryotic species based on their evolutionary characteristics.

## File Formats

### FASTA Files

- Standard FASTA format with protein sequences
- Header information includes accession numbers, protein families, and organism taxonomy
- All sequences have been filtered and preprocessed

### PDB Files

- Protein Data Bank (PDB) format containing 3D atomic coordinates
- Standard ATOM records with positions, element types, and connectivity
- Each structure corresponds to its respective sequence in the FASTA files
- **Prediction Tool**: All PDB structures in this dataset are predicted using ESMFold. 

## Dataset Statistics

| Category  | FASTA Files | PDB Files  | Total Sequences |
| --------- | ----------- | ---------- | --------------- |
| Positive  | 1           | 1,421      | 1,421           |
| Negative  | 1           | 52,824     | 52,824          |
| **Total** | **2**       | **54,245** | **54,245**      |

## Applications

This dataset is designed for:

- Anti-CRISPR protein prediction and classification
- Machine learning model training and benchmarking
- Structure-function relationship analysis
- Protein sequence-structure studies
- CRISPR-Cas system inhibition research

## Usage

When using this dataset, please cite the appropriate source and acknowledge that this is a benchmark dataset for anti-CRISPR protein research.

## Notes

- All protein structures have been preprocessed and standardized
- The dataset maintains a 1:1 mapping between FASTA sequences and PDB structures
- Negative samples have been curated to provide challenging non-anti-CRISPR examples
- Data has been filtered for quality and consistency