File size: 2,152 Bytes
acbf15a
 
 
 
 
 
 
 
4248c18
acbf15a
 
 
 
 
 
4a7494d
acbf15a
 
4248c18
acbf15a
 
4248c18
acbf15a
4a7494d
 
acbf15a
 
 
 
 
 
 
 
 
 
237032f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
111202a
 
bd0fed0
111202a
de1a84e
 
 
 
 
 
 
 
 
237032f
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
---
dataset_info:
  features:
  - name: A
    dtype: string
  - name: B
    dtype: string
  - name: labels
    dtype: int8
  - name: SeqA
    dtype: string
  - name: SeqB
    dtype: string
  splits:
  - name: train
    num_bytes: 271796192
    num_examples: 163192
  - name: valid
    num_bytes: 73318294
    num_examples: 59260
  - name: test
    num_bytes: 57573817
    num_examples: 52048
  download_size: 291089780
  dataset_size: 402688303
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: valid
    path: data/valid-*
  - split: test
    path: data/test-*
---

# Leakage-free "gold" standard PPI dataset

From [Bernett, et al](https://figshare.com/articles/dataset/PPI_prediction_from_sequence_gold_standard_dataset/21591618), found in 

Cracking the black box of deep sequence-based protein–protein interaction prediction
* [paper](https://academic.oup.com/bib/article/25/2/bbae076/7621029)
* [code](https://github.com/daisybio/data-leakage-ppi-prediction)

and 

Deep learning models for unbiased sequence-based PPI prediction plateau at an accuracy of 0.65
* [paper](https://www.biorxiv.org/content/10.1101/2025.01.23.634454v1)
* [code](https://github.com/daisybio/PPI_prediction_study)

## Description
This is a balanced binary protein-protein interaction dataset with positives from [HIPPIE](https://cbdm-01.zdv.uni-mainz.de/~mschaefer/hippie/) and paritioned with [KaHIP](https://kahip.github.io/). There are no sequence overlaps in splits, furthermore, they are split based on a maximum CD-HIT of 40% pairwise sequence similarity. Node degree bias was also reduced.

Note:

[Q96PU5](https://www.uniprot.org/uniprotkb/Q96PU5) is not located in the provided SwissProt fasta file but is used in the train split several times. We added this before data processing so no rows were dropped.

## Example use
```python
def get_ppi_data():
    data = load_dataset("Synthyra/bernett_gold_ppi").shuffle(seed=42)
    data = data.remove_columns(['A', 'B'])
    return data['train'], data['valid'], data['test']
```

## Please cite
Please cite their figshare dataset and papers if you use this dataset.