---
dataset_info:
  features:
  - name: pdb_id
    dtype: string
  - name: labels
    dtype: float64
  - name: SeqA
    dtype: string
  - name: SeqB
    dtype: string
  - name: source
    dtype: string
  - name: kd
    dtype: float64
  splits:
  - name: train
    num_bytes: 5998837
    num_examples: 11076
  download_size: 1697045
  dataset_size: 5998837
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---


# Affinity dataset from [Bindwell](https://github.com/Bindwell/APPT)

Natively fairly nonredundant via indel sequence similarity calculations.
<img src="https://cdn-uploads.huggingface.co/production/uploads/62f2bd3bdb7cbd214b658c48/o7iHaBUhk6fwZusSusls9.png" width="600">


Note:

There are two NaNs in the `protein2_sequence` column of the original dataset - we removed these rows.

## Example use

```python
from datasets import load_dataset

def get_affinity_data():
    data = load_dataset("Synthyra/ProteinProteinAffinity", split="train")
    data = data.remove_columns(['pdb_id', 'source', 'kd'])
    data = data.shuffle(seed=11)
    data = data.train_test_split(test_size=1000, seed=22)
    valid = data['test']
    train = data['train']
    valid = valid.train_test_split(test_size=500, seed=33)
    test = valid['test']
    valid = valid['train']
    return train, valid, test
```

## Please cite
Please cite Bindwell's work if you use this dataset.