# Dataset Card for Contact Prediction Dataset for RAGProtein

### Dataset Summary

Contact map prediction aims to determine whether two residues, $i$ and $j$, are in contact or not, based on their distance with a certain threshold ($<$8 Angstrom). This task is an important part of the early Alphafold version for structural prediction. 

## Dataset Structure

### Data Instances

For each instance, there is a string of the protein sequences, a sequence for the contact labels. Each of the sub-labels "[2, 3]" indicates the 3rd residue are in contact with the 4th residue (start from index 0). See the [Contact map prediction dataset viewer](https://huggingface.co/datasets/Bo1015/contact_prediction_binary/viewer/default/test) to explore more examples.

```
{'seq':'QNLLKNLAASLGRKPFVADKQGVYRLTIDKHLVMLAPHGSELVLRTPIDAPMLREGNNVNVTLLRSLMQQALAWAKRYPQTLVLDDCGQLVLEARLRLQELDTHGLQEVINKQLALLEHLIPQLTP'
'label': [ [ 0, 0 ], [ 0, 1 ], [ 1, 1 ], [ 1, 2 ], [ 1, 3 ], [ 1, 101 ], [ 2, 2 ], [ 2, 3 ], [ 2, 4 ], [ 3, 3 ], [ 3, 4 ], [ 3, 5 ], [ 3, 99 ], [ 3, 100 ], [ 3, 101 ], [ 4, 4 ], [ 4, 5 ], [ 4, 53 ], ...], 'msa': 'QNLLKNLAASLGRKPFVADKQGVYRLTIDKHLVMLAPHGSELVLRTPIDAPMLREGNNVNVTLLRSLMQQALAWAKRYPQTLVLDDCGQLVLEARLRLQELDTHGLQEVINKQLALLEHLIPQLTP|QNLLKNLAASLGRKPFVADKQGVYRLTIDKHLVMLAPHGSELVLRTPIDAPMLREGNNVNVTLLRSLMQQALAWAKRYPQTLVLDDCGQLVLEARLRLQELDTHGLQEVINKQLALLEHLIPQLTP...', 'str_emb': [seq_len, 384]}
```

The average  for the `seq` and the `label` are provided below:

| Feature    | Mean Count |
| ---------- | ---------------- |
| seq    | 249        |
| label | 1,500             |

### Data Fields

- `seq`: a string containing the protein sequence
- `label`: a string containing the contact label of each residue pair.
- `msa`: "|" seperated MSA sequences
- `str_emb`: AIDO.StructureTokenizer generated structure embedding from AF2 predicted structures

### Data Splits

The contact map prediction dataset has 3 splits: _train_, _validation_, and _test_. Below are the statistics of the dataset.

| Dataset Split | Number of Instances in Split                |
| ------------- | ------------------------------------------- |
| Train         | 12,041                              |
| Validation    | 1,505                                      |
| Test          | 1,505                                      |

### Source Data

#### Initial Data Collection and Normalization

The [trRosetta dataset](https://www.pnas.org/doi/10.1073/pnas.1914677117) is employed as the initilized dataset. 

### Licensing Information

The dataset is released under the [Apache-2.0 License](http://www.apache.org/licenses/LICENSE-2.0). 

### Processed data collection 

Single sequence data are collected from this paper:

```
@misc{chen2024xtrimopglm,
  title={xTrimoPGLM: unified 100B-scale pre-trained transformer for deciphering the language of protein},
  author={Chen, Bo and Cheng, Xingyi and Li, Pan and Geng, Yangli-ao and Gong, Jing and Li, Shen and Bei, Zhilei and Tan, Xu and Wang, Boyan and Zeng, Xin and others},
  year={2024},
  eprint={2401.06199},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  note={arXiv preprint arXiv:2401.06199}
}
```