---
dataset_info:
- config_name: clustered
  features:
  - name: text
    dtype: string
  - name: labels
    dtype: string
  splits:
  - name: train
    num_bytes: 392413559
    num_examples: 1297604
  - name: validation
    num_bytes: 3452474
    num_examples: 11315
  - name: test
    num_bytes: 3463001
    num_examples: 11296
  - name: held_out
    num_bytes: 1573513
    num_examples: 5670
  download_size: 153375375
  dataset_size: 400902547
- config_name: large
  features:
  - name: text
    dtype: string
  - name: labels
    dtype: string
  splits:
  - name: train
    num_bytes: 408033438.0423948
    num_examples: 1366992
  - name: validation
    num_bytes: 4163338.478802599
    num_examples: 13948
  - name: test
    num_bytes: 4163338.478802599
    num_examples: 13948
  - name: held_out
    num_bytes: 1573513
    num_examples: 5670
  download_size: 165141976
  dataset_size: 417933628.00000006
- config_name: standard
  features:
  - name: text
    dtype: string
  - name: labels
    dtype: string
  splits:
  - name: train
    num_bytes: 413375215.0295364
    num_examples: 1384888
  - name: validation
    num_bytes: 1492449.9852317893
    num_examples: 5000
  - name: test
    num_bytes: 1492449.9852317893
    num_examples: 5000
  - name: held_out
    num_bytes: 1573513
    num_examples: 5670
  download_size: 165141561
  dataset_size: 417933628.00000006
configs:
- config_name: clustered
  data_files:
  - split: train
    path: clustered/train-*
  - split: validation
    path: clustered/validation-*
  - split: test
    path: clustered/test-*
  - split: held_out
    path: clustered/held_out-*
- config_name: large
  data_files:
  - split: train
    path: large/train-*
  - split: validation
    path: large/validation-*
  - split: test
    path: large/test-*
  - split: held_out
    path: large/held_out-*
- config_name: standard
  data_files:
  - split: train
    path: standard/train-*
  - split: validation
    path: standard/validation-*
  - split: test
    path: standard/test-*
  - split: held_out
    path: standard/held_out-*
license: mit
task_categories:
- text2text-generation
tags:
- chemistry
- medical
- targetd-protein-degradation
- protac
---

# ✂️ PROTAC-Splitter Dataset ✂️

This Hugging-Face dataset contains the data for training and evaluating the Tranformer-based PROTAC-Splitter model.

If you find this dataset useful or want to know more, please consider reading and citing the following work:

```
@article{Ribes2025PROTACSplitter,
  title   = {PROTAC‐Splitter: A Machine Learning Framework for Automated Identification of PROTAC Substructures},
  author  = {Stefano Ribes and Ranxuan Zhang and Télio Cropsal and Anders Källberg and Christian Tyrchan and Eva Nittinger and Rocío Mercado},
  journal = {ChemRxiv},
  year    = {2025},
  month   = {Jul},
  day     = {08},
  doi     = {10.26434/chemrxiv-2025-bn1nv},
  url     = {https://chemrxiv.org/engage/chemrxiv/article-details/686670983ba0887c33677fc8},
  license = {CC BY 4.0}
}
```

Additional information on the models and data can also be found at this Zenodo link: [https://zenodo.org/records/15797310](https://zenodo.org/records/15797310)

## GitHub Repository 📝

The code for training and evaluation the PROTAC-Splitter models can be found at: [https://github.com/ribesstefano/PROTAC-Splitter](https://github.com/ribesstefano/PROTAC-Splitter)